deepseek-harness/apps/cli/tests/web-agent-presets.e2e.ts

962 lines
44 KiB
TypeScript
Raw Normal View History

import { randomUUID } from 'node:crypto'
import { mkdir, mkdtemp, readFile, stat, symlink, writeFile } from 'node:fs/promises'
feat(agent-presets): make the default preset a user setting `config.default` becomes the composition base of an `agent-presets` settings namespace, so the user document layers over the deployment's engineering default and a person can change which preset new sessions get without a restart. The value is read per resolution rather than snapshotted: a hot-reloaded document takes effect on the next session created, and every running session stays on the preset it was composed from — which is the same rule the session-header guard enforces from the other side. `resolve()` read `config.default` directly, which would have made the whole setting inert; it now goes through `defaultId` like every other caller. The write-protection test is rewritten against a temp profile root. It was passing vacuously: the un-overridden Loader REWRITES the composition it read — stamping `disabled: true` onto the self-disposing row — so the committed fixture had been mutated by the very run that proved the bug, and every later run compared against the damaged file and passed. Building the preset in a temp directory makes the assertion immune to its own failure mode, and it now fails with a visible `+ disabled: true` when the override is removed. Review follow-ups on this layer. The exported schema is `AgentPresetSettingsSchema`, symmetric with the `AgentPresetSettings` interface it resolves and self-describing at an import site. The `session.create` JSDoc promised "the deployment's default preset" for an omitted `agentPreset`, which this layer makes false — it now names the effective default. The constructor records why it does not use `installSettingsSection`: that helper re-judges what a consumer DERIVED across attach and detach, and nothing here is derived. The provider-unload test disposes the fiber `ctx.plugin()` handed back instead of reaching into `ctx.reflect.store`, and the write-protection wait says why slack is the right shape for an absence assertion. The real composition covers the layering too. `apps/cli` boots the shipped `cordis.yml`, stores `agent-presets.default`, and asserts an unnamed session composes from it — the package suite proves the layering against a hand-built context, this proves the roster and the settings provider are wired to each other. That test also pins the settings row at a temp file: it defaulted to `$DSH_HOME/settings.yaml`, so a developer's own stored default decided the outcome of a file whose whole point is that only the shipped root does. The Agent Note records the per-resolution read and its correspondence with the session header, and the vacuous-test finding above.
2026-08-04 00:07:11 +08:00
import { tmpdir } from 'node:os'
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
import { fileURLToPath } from 'node:url'
fix(web): return every service a host row injects, and run the browser lane The browser e2e lane had been failing wholesale since this stack moved the agent plane into presets, and nothing caught it: 34 of 48 files. Two of the causes are product defects, not test breakage. `bashEnv` goes back to the host plane. `apps/cli/src/web.ts` injects it to publish `DSH_WEB_URL`/`DSH_WEB_MODE`, so the earlier note that "nothing outside the agent plane injects bashEnv" was simply wrong — behind a preset's `shell` realm those variables reached no shell at all, and a `dsh web` agent could not find the address of its own interface. This is the same criterion that returned `subagents`: a host row that injects a service resolves before any session exists and has no agent to key by, so the service is host-plane. `tool-bash` consumes the host registry from inside the preset, which works because an agent context chains to the host; only the reverse is invisible. `tool-subagent-report` goes back with it. It is not a tool this agent calls: it registers a continuable SETUP on the host `subagents` singleton, and that list is not scope-aware. One copy per mounted preset meant every child was handed `report` once per live session, so the second registration threw and a cold subagent resume failed with `subagent-not-resumable` — a diagnostic three layers removed from the cause. The lane's own composition facts follow. Skill roots resolve inside a preset now, a subtree include patches cannot reach, so the scaffold pins the roots' documented environment fallback for its whole lifetime rather than for the boot — presets mount per session. Without it the developer's real `~/.dsh/skills` enters replay requests and goldens while CI sees none. The `apps/cli` composition test pins `storage-json` for the same reason: unpinned it wrote, and then read back, the developer's own `~/.dsh/storages/`. Three tests now address through an agent what they used to read off the root context, because that is where the thing lives: the tool catalog, the skill registry, and the token meter. The seeded-history projection baseline asserts the opposite of what it did — a detached session yields a preset-plane projection only from a durable checkpoint written while it was live, and this seed was written straight to persistence and never ran. Goldens re-recorded for the hero's preset chip and the settings nav entry.
2026-08-06 19:14:13 +08:00
import { dirname, join } from 'node:path'
build(vendor): rescope the vendored Cordis packages into @deepseek-ai Machine-produced by `pnpm run rescope-vendor --apply` plus the regeneration it prints: `pnpm install` for the lockfile, `pnpm run gen-third-party-notices`, `verify-translation-pairing --write` for the touched bilingual pairs, `gen-doc-graphs`, and one typert snapshot whose ids embed character offsets. `pnpm run rescope-vendor --check` verifies the result. Renames nine vendored packages (cordis, cosmokit, schemastery and the six @cordisjs plugins) and every reference that resolves them: manifest names and dependency keys, module specifiers including declare-module merges, cordis.yml plugin names, tsconfig paths, every Markdown fence, and `docs/` prose. Directory names, upstream versions, and dependency ranges are unchanged, so vendor/README.md still reads as an upstream snapshot; its manifest table gains an upstream-name column so THIRD_PARTY_NOTICES keeps MIT attribution pointed at each fork's origin. The tutorial tier follows the rename end to end: its yaml fences named plugins the Loader can no longer resolve, its `ts ignore-check` fences disagreed with the compiled fences beside them, and its prose quoted both. The contracts that told readers to keep upstream names — the root convention and the vendoring cookbook's tree comment and manifest invariant — now say to rescope instead. Two rules read `@deepseek-ai/` as "another workspace plugin": the client bundle purity gate now names the vendored libraries a browser bundle inlines, and the files where a bare `cordis` is an agent-preset id keep that product data.
2026-08-10 22:04:06 +08:00
import { Context } from '@deepseek-ai/cordis'
import { boot, healProfilesModuleFallback, loadOverlayPatches, loadProfile } from '@deepseek-ai/dsh-app-boot'
import { provideCmdline } from '@deepseek-ai/dsh-cmdline'
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
import { SessionId } from '@deepseek-ai/dsh-session'
import type { Agent } from '@deepseek-ai/dsh-agent'
build(vendor): rescope the vendored Cordis packages into @deepseek-ai Machine-produced by `pnpm run rescope-vendor --apply` plus the regeneration it prints: `pnpm install` for the lockfile, `pnpm run gen-third-party-notices`, `verify-translation-pairing --write` for the touched bilingual pairs, `gen-doc-graphs`, and one typert snapshot whose ids embed character offsets. `pnpm run rescope-vendor --check` verifies the result. Renames nine vendored packages (cordis, cosmokit, schemastery and the six @cordisjs plugins) and every reference that resolves them: manifest names and dependency keys, module specifiers including declare-module merges, cordis.yml plugin names, tsconfig paths, every Markdown fence, and `docs/` prose. Directory names, upstream versions, and dependency ranges are unchanged, so vendor/README.md still reads as an upstream snapshot; its manifest table gains an upstream-name column so THIRD_PARTY_NOTICES keeps MIT attribution pointed at each fork's origin. The tutorial tier follows the rename end to end: its yaml fences named plugins the Loader can no longer resolve, its `ts ignore-check` fences disagreed with the compiled fences beside them, and its prose quoted both. The contracts that told readers to keep upstream names — the root convention and the vendoring cookbook's tree comment and manifest invariant — now say to rescope instead. Two rules read `@deepseek-ai/` as "another workspace plugin": the client bundle purity gate now names the vendored libraries a browser bundle inlines, and the files where a bare `cordis` is an agent-preset id keep that product data.
2026-08-10 22:04:06 +08:00
import type { PatchOptions } from '@deepseek-ai/cordis-plugin-include'
import { afterAll, beforeAll, describe, expect, it, vi } from 'vitest'
feat(agent-presets): make the default preset a user setting `config.default` becomes the composition base of an `agent-presets` settings namespace, so the user document layers over the deployment's engineering default and a person can change which preset new sessions get without a restart. The value is read per resolution rather than snapshotted: a hot-reloaded document takes effect on the next session created, and every running session stays on the preset it was composed from — which is the same rule the session-header guard enforces from the other side. `resolve()` read `config.default` directly, which would have made the whole setting inert; it now goes through `defaultId` like every other caller. The write-protection test is rewritten against a temp profile root. It was passing vacuously: the un-overridden Loader REWRITES the composition it read — stamping `disabled: true` onto the self-disposing row — so the committed fixture had been mutated by the very run that proved the bug, and every later run compared against the damaged file and passed. Building the preset in a temp directory makes the assertion immune to its own failure mode, and it now fails with a visible `+ disabled: true` when the override is removed. Review follow-ups on this layer. The exported schema is `AgentPresetSettingsSchema`, symmetric with the `AgentPresetSettings` interface it resolves and self-describing at an import site. The `session.create` JSDoc promised "the deployment's default preset" for an omitted `agentPreset`, which this layer makes false — it now names the effective default. The constructor records why it does not use `installSettingsSection`: that helper re-judges what a consumer DERIVED across attach and detach, and nothing here is derived. The provider-unload test disposes the fiber `ctx.plugin()` handed back instead of reaching into `ctx.reflect.store`, and the write-protection wait says why slack is the right shape for an absence assertion. The real composition covers the layering too. `apps/cli` boots the shipped `cordis.yml`, stores `agent-presets.default`, and asserts an unnamed session composes from it — the package suite proves the layering against a hand-built context, this proves the roster and the settings provider are wired to each other. That test also pins the settings row at a temp file: it defaulted to `$DSH_HOME/settings.yaml`, so a developer's own stored default decided the outcome of a file whose whole point is that only the shipped root does. The Agent Note records the per-resolution read and its correspondence with the session header, and the vacuous-test finding above.
2026-08-04 00:07:11 +08:00
import { settingsNamespace } from '@deepseek-ai/dsh-settings'
import { SUBAGENT_MODEL_SELECTION_SETTINGS_NAMESPACE } from '@deepseek-ai/dsh-tool-subagent/model-selection-settings'
import { SETTINGS_NAMESPACE, SHIPPED_PRESET_ROOT } from '@deepseek-ai/dsh-agent-presets'
fix(subagent): compose children from their parent's preset Tool and prompt-section visibility is inherited along dsh-scope's parent chain, and an agent's scope key is minted with no parent. Per-session agent presets moved every model-facing row onto the agent plane and made AgentPresets.mount() the one thing that binds that link, from the api-proxy's session create, resume, and fork paths. The two in-process subagent drivers installed only the per-child persona and tool filter, so a child's scope chain had length one and its registry view resolved the global layer alone — which is empty wherever a preset roster is composed. One-shot children reached the model with no tools, continuable ones with only the host-plane `report`, and neither carried its parent's persona, workspace context, or skill catalog. AgentPresets.composeFrom() joins one agent to the standing composition another already runs on. It is a bind, not a mount: the child gets its parent's exact generation, so a composition edited since the parent started cannot fork it onto another one, and it is synchronous, which is what lets a child creation window use it. applyChildComposition() now takes the parent and performs the join first, making a child composed without it unrepresentable at the call sites. childSessionMeta() records the joined id so a cold read rebuilds the composition the child actually ran under. The audit that followed found two api-proxy readers on the wrong authority: presenterScopeFor() and the live-agent branch of assertPresetUnchanged() both read header.agentPreset, which goes stale the moment a blank session switches preset. A switched session's cold transcript resolved presenters in the older composition's layer and silently degraded to generic cards, and the gateway refused to adopt a live session under the preset it actually runs while accepting the one it left. Both now resolve through resolveSessionPreset(), matching the resume branch fifteen lines above. The owning architecture Agent Note carried the stale claim that the header records what a session runs; it is corrected to name the header/log pair and its three readers. Fixes #2165
2026-08-10 17:46:34 +08:00
import { applyChildComposition, childSessionMeta } from '@deepseek-ai/dsh-subagent'
import { CallId } from '@deepseek-ai/dsh-llm'
import type {} from '@deepseek-ai/dsh-compaction-basic'
import type {} from '@deepseek-ai/dsh-skill'
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
import type {} from '@deepseek-ai/dsh-tools'
fix(preset): keep the token meter host-plane and name unjoined agents Moving the agent plane behind presets left two readers on the wrong side of the host/agent line. `dsh-token-meter` was disabled on the host and mounted inside each preset's `compaction` realm, but its three projection units register into the process-wide `sessionProjections` table. A unit registered from one preset answers for every session, so whether a `minimal` session showed a context meter depended on whether some other session had mounted `standard` since boot, and a process that only ever ran `minimal` showed none. The meter takes no configuration, keys every fold by Session, and registers no tool or prompt section, so it returns to the host composition and leaves the presets' `isolate` map; the realm and `compact-basic` stay, because what a preset chooses is whether its agent compacts, not whether its tokens are counted. Nothing named an agent that joined no preset. The join is a scope-parent link, and without it the tools, prompt-section, and skill views resolve the empty global layer: the agent publishes, the turn runs, and the model receives nothing. `AgentPresets` now logs one warning per such agent while a roster is configured, and the invariant companion fails outright — at `system-prompt/assemble` rather than at publication, because an unjoined agent is legal until it addresses a model and `recompose` binds exactly such an agent. The warning stays advisory: a synchronous `agent/created` throw vetoes publication, and the ACP bridge, SDK server, and headless bundle all create an unjoined agent today. Three limits are recorded rather than fixed: projection key presence is not a per-session capability signal, a superseded standing generation is never reclaimed, and a `cordis_mount` temporary plugin belongs to the composition rather than the session that mounted it. Fixes #2203
2026-08-10 22:36:06 +08:00
// Type-only: resolves `ctx.get('sessionProjections')` and `ctx.get('tokenMeter')`.
import type {} from '@deepseek-ai/dsh-session-projection'
import type {} from '@deepseek-ai/dsh-token-meter'
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
const REPO_ROOT = fileURLToPath(new URL('../../..', import.meta.url))
/** The shipped Web surface: the dsh-base and dsh-web-app bundle patches over an empty preset root. */
const BASE_PATCH = join(REPO_ROOT, 'packages/bundle/base/cordis.patch.yml')
const WEB_PATCH = join(REPO_ROOT, 'packages/bundle/web-app/cordis.patch.yml')
const CODEX_PACKAGE_DIR = join(REPO_ROOT, 'packages/subagent/subagent-codex')
const CLAUDE_CODE_PACKAGE_DIR = join(REPO_ROOT, 'packages/subagent/subagent-claude-code')
/** The installation anchor whose dependency surface the preset module fallback mirrors. */
const INSTALL_ANCHOR = join(REPO_ROOT, 'apps/cli/package.json')
const MINIMAL_PROMPT = 'You are a helpful software engineer assistant.'
const MINIMAL_BASH_DESCRIPTION = `Run commands in a bash shell
* When invoking this tool, the contents of the "command" parameter does NOT need to be XML-escaped.
* You don't have access to the internet via this tool.
* You do have access to a mirror of common linux and python packages via apt and pip.
* State is persistent across command calls and discussions with the user.
* To inspect a particular line range of a file, e.g. lines 10-25, try 'sed -n 10,25p /path/to/the/file'.
* Please avoid commands that may produce a very large amount of output.
* Please run long lived commands in the background, e.g. 'sleep 10 &' or start a server in the background.`
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
/**
* Boot the shipped Web composition, minus the rows that would bind a port,
* touch the network, or write outside the test. Everything that decides an
* agent's capabilities is the real thing, including both shipped presets.
*/
async function bootWeb(
settingsFile: string,
extra: PatchOptions[] = [],
profilePackages: readonly string[] = [],
profileBundles?: readonly string[],
): Promise<Context> {
fix(web): return every service a host row injects, and run the browser lane The browser e2e lane had been failing wholesale since this stack moved the agent plane into presets, and nothing caught it: 34 of 48 files. Two of the causes are product defects, not test breakage. `bashEnv` goes back to the host plane. `apps/cli/src/web.ts` injects it to publish `DSH_WEB_URL`/`DSH_WEB_MODE`, so the earlier note that "nothing outside the agent plane injects bashEnv" was simply wrong — behind a preset's `shell` realm those variables reached no shell at all, and a `dsh web` agent could not find the address of its own interface. This is the same criterion that returned `subagents`: a host row that injects a service resolves before any session exists and has no agent to key by, so the service is host-plane. `tool-bash` consumes the host registry from inside the preset, which works because an agent context chains to the host; only the reverse is invisible. `tool-subagent-report` goes back with it. It is not a tool this agent calls: it registers a continuable SETUP on the host `subagents` singleton, and that list is not scope-aware. One copy per mounted preset meant every child was handed `report` once per live session, so the second registration threw and a cold subagent resume failed with `subagent-not-resumable` — a diagnostic three layers removed from the cause. The lane's own composition facts follow. Skill roots resolve inside a preset now, a subtree include patches cannot reach, so the scaffold pins the roots' documented environment fallback for its whole lifetime rather than for the boot — presets mount per session. Without it the developer's real `~/.dsh/skills` enters replay requests and goldens while CI sees none. The `apps/cli` composition test pins `storage-json` for the same reason: unpinned it wrote, and then read back, the developer's own `~/.dsh/storages/`. Three tests now address through an agent what they used to read off the root context, because that is where the thing lives: the tool catalog, the skill registry, and the token meter. The seeded-history projection baseline asserts the opposite of what it did — a detached session yields a preset-plane projection only from a durable checkpoint written while it was live, and this seed was written straight to persistence and never ran. Goldens re-recorded for the hero's preset chip and the settings nav entry.
2026-08-06 19:14:13 +08:00
const storageRoot = join(dirname(settingsFile), 'storages')
const overrides: PatchOptions[] = [
feat(agent-presets): make the default preset a user setting `config.default` becomes the composition base of an `agent-presets` settings namespace, so the user document layers over the deployment's engineering default and a person can change which preset new sessions get without a restart. The value is read per resolution rather than snapshotted: a hot-reloaded document takes effect on the next session created, and every running session stays on the preset it was composed from — which is the same rule the session-header guard enforces from the other side. `resolve()` read `config.default` directly, which would have made the whole setting inert; it now goes through `defaultId` like every other caller. The write-protection test is rewritten against a temp profile root. It was passing vacuously: the un-overridden Loader REWRITES the composition it read — stamping `disabled: true` onto the self-disposing row — so the committed fixture had been mutated by the very run that proved the bug, and every later run compared against the damaged file and passed. Building the preset in a temp directory makes the assertion immune to its own failure mode, and it now fails with a visible `+ disabled: true` when the override is removed. Review follow-ups on this layer. The exported schema is `AgentPresetSettingsSchema`, symmetric with the `AgentPresetSettings` interface it resolves and self-describing at an import site. The `session.create` JSDoc promised "the deployment's default preset" for an omitted `agentPreset`, which this layer makes false — it now names the effective default. The constructor records why it does not use `installSettingsSection`: that helper re-judges what a consumer DERIVED across attach and detach, and nothing here is derived. The provider-unload test disposes the fiber `ctx.plugin()` handed back instead of reaching into `ctx.reflect.store`, and the write-protection wait says why slack is the right shape for an absence assertion. The real composition covers the layering too. `apps/cli` boots the shipped `cordis.yml`, stores `agent-presets.default`, and asserts an unnamed session composes from it — the package suite proves the layering against a hand-built context, this proves the roster and the settings provider are wired to each other. That test also pins the settings row at a temp file: it defaulted to `$DSH_HOME/settings.yaml`, so a developer's own stored default decided the outcome of a file whose whole point is that only the shipped root does. The Agent Note records the per-resolution read and its correspondence with the session header, and the vacuous-test finding above.
2026-08-04 00:07:11 +08:00
// The settings row defaults to `$DSH_HOME/settings.yaml`. Left alone it
// reads the developer's own document — and since the default preset is a
// setting, a stored `agent-presets.default` would decide this file's
// outcome. Point it at a temp file for the same reason the roster row
// below pins `includeUserRoot` off.
feat(agent-presets): make the default preset a user setting `config.default` becomes the composition base of an `agent-presets` settings namespace, so the user document layers over the deployment's engineering default and a person can change which preset new sessions get without a restart. The value is read per resolution rather than snapshotted: a hot-reloaded document takes effect on the next session created, and every running session stays on the preset it was composed from — which is the same rule the session-header guard enforces from the other side. `resolve()` read `config.default` directly, which would have made the whole setting inert; it now goes through `defaultId` like every other caller. The write-protection test is rewritten against a temp profile root. It was passing vacuously: the un-overridden Loader REWRITES the composition it read — stamping `disabled: true` onto the self-disposing row — so the committed fixture had been mutated by the very run that proved the bug, and every later run compared against the damaged file and passed. Building the preset in a temp directory makes the assertion immune to its own failure mode, and it now fails with a visible `+ disabled: true` when the override is removed. Review follow-ups on this layer. The exported schema is `AgentPresetSettingsSchema`, symmetric with the `AgentPresetSettings` interface it resolves and self-describing at an import site. The `session.create` JSDoc promised "the deployment's default preset" for an omitted `agentPreset`, which this layer makes false — it now names the effective default. The constructor records why it does not use `installSettingsSection`: that helper re-judges what a consumer DERIVED across attach and detach, and nothing here is derived. The provider-unload test disposes the fiber `ctx.plugin()` handed back instead of reaching into `ctx.reflect.store`, and the write-protection wait says why slack is the right shape for an absence assertion. The real composition covers the layering too. `apps/cli` boots the shipped `cordis.yml`, stores `agent-presets.default`, and asserts an unnamed session composes from it — the package suite proves the layering against a hand-built context, this proves the roster and the settings provider are wired to each other. That test also pins the settings row at a temp file: it defaulted to `$DSH_HOME/settings.yaml`, so a developer's own stored default decided the outcome of a file whose whole point is that only the shipped root does. The Agent Note records the per-resolution read and its correspondence with the session header, and the vacuous-test finding above.
2026-08-04 00:07:11 +08:00
{ id: 'settings', config: { path: settingsFile, watch: false } },
fix(web): return every service a host row injects, and run the browser lane The browser e2e lane had been failing wholesale since this stack moved the agent plane into presets, and nothing caught it: 34 of 48 files. Two of the causes are product defects, not test breakage. `bashEnv` goes back to the host plane. `apps/cli/src/web.ts` injects it to publish `DSH_WEB_URL`/`DSH_WEB_MODE`, so the earlier note that "nothing outside the agent plane injects bashEnv" was simply wrong — behind a preset's `shell` realm those variables reached no shell at all, and a `dsh web` agent could not find the address of its own interface. This is the same criterion that returned `subagents`: a host row that injects a service resolves before any session exists and has no agent to key by, so the service is host-plane. `tool-bash` consumes the host registry from inside the preset, which works because an agent context chains to the host; only the reverse is invisible. `tool-subagent-report` goes back with it. It is not a tool this agent calls: it registers a continuable SETUP on the host `subagents` singleton, and that list is not scope-aware. One copy per mounted preset meant every child was handed `report` once per live session, so the second registration threw and a cold subagent resume failed with `subagent-not-resumable` — a diagnostic three layers removed from the cause. The lane's own composition facts follow. Skill roots resolve inside a preset now, a subtree include patches cannot reach, so the scaffold pins the roots' documented environment fallback for its whole lifetime rather than for the boot — presets mount per session. Without it the developer's real `~/.dsh/skills` enters replay requests and goldens while CI sees none. The `apps/cli` composition test pins `storage-json` for the same reason: unpinned it wrote, and then read back, the developer's own `~/.dsh/storages/`. Three tests now address through an agent what they used to read off the root context, because that is where the thing lives: the tool catalog, the skill registry, and the token meter. The seeded-history projection baseline asserts the opposite of what it did — a detached session yields a preset-plane projection only from a durable checkpoint written while it was live, and this seed was written straight to persistence and never ran. Goldens re-recorded for the hero's preset chip and the settings nav entry.
2026-08-06 19:14:13 +08:00
// storage-json's root is anchored to the real $DSH_HOME. Unpinned, this
// file writes the developer's own `~/.dsh/storages/` — and then reads it
// back on the next run, so a stored document from any other build decides
// this test's boot. Same reason the settings row above is pinned.
{ id: 'storage-json', config: { root: storageRoot } },
// Host rows with side effects outside this process: a bound port, a served
// asset tree, a telemetry exporter. `api-gateway` and `directory-picker`
// stay ENABLED on purpose — the api-proxy is the host row that injects
// `subagents`, `workspace`, and the rest of the agent plane, so disabling
// it would hide exactly the breakage this file exists to catch: a service
// moved into the presets that a host row still waits for. The boot audit
// is that assertion.
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
{ id: 'webserver', disabled: true },
// The web bundle's runtime row injects `webServer`, so it cannot
// activate without the bound port disabled above. It owns dist serving
// and the URL prompt line — surface glue, not anything that decides an
// agent's capabilities, which is all this file asserts.
{ id: 'web-runtime', disabled: true },
{ id: 'session-telemetry-otel', disabled: true },
// A deployment-level skill on the host registry's GLOBAL layer — the same
// registration shape a repository plugin's skill root uses. The layered
// skills test below proves it reaches preset-composed agents.
{ id: 'skill-badge', disabled: false },
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
{ id: 'modules', disabled: true },
{ id: 'connection', disabled: true },
// The always-on reload chain waits for the browser roster and bound port
// disabled above.
{ id: 'client-hmr', disabled: true },
// The shipped `-auto` chooser resolves its interaction from a running
// host and so waits for the webserver disabled above; the browse variant
// supplies `directoryPicker` without one.
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
{ id: 'directory-picker', disabled: true },
{ insert: [
{ id: 'directory-picker-browse', name: '@deepseek-ai/dsh-host-directory-picker-browse' },
{ id: 'ui-directory-picker-browse', name: '@deepseek-ai/dsh-client-ui-directory-picker-browse' },
] },
// Pin the roster away from the developer's machine: `includeUserRoot`
// false keeps `~/.dsh/.agent-presets` from changing a test's outcome.
feat(agent-presets): make the default preset a user setting `config.default` becomes the composition base of an `agent-presets` settings namespace, so the user document layers over the deployment's engineering default and a person can change which preset new sessions get without a restart. The value is read per resolution rather than snapshotted: a hot-reloaded document takes effect on the next session created, and every running session stays on the preset it was composed from — which is the same rule the session-header guard enforces from the other side. `resolve()` read `config.default` directly, which would have made the whole setting inert; it now goes through `defaultId` like every other caller. The write-protection test is rewritten against a temp profile root. It was passing vacuously: the un-overridden Loader REWRITES the composition it read — stamping `disabled: true` onto the self-disposing row — so the committed fixture had been mutated by the very run that proved the bug, and every later run compared against the damaged file and passed. Building the preset in a temp directory makes the assertion immune to its own failure mode, and it now fails with a visible `+ disabled: true` when the override is removed. Review follow-ups on this layer. The exported schema is `AgentPresetSettingsSchema`, symmetric with the `AgentPresetSettings` interface it resolves and self-describing at an import site. The `session.create` JSDoc promised "the deployment's default preset" for an omitted `agentPreset`, which this layer makes false — it now names the effective default. The constructor records why it does not use `installSettingsSection`: that helper re-judges what a consumer DERIVED across attach and detach, and nothing here is derived. The provider-unload test disposes the fiber `ctx.plugin()` handed back instead of reaching into `ctx.reflect.store`, and the write-protection wait says why slack is the right shape for an absence assertion. The real composition covers the layering too. `apps/cli` boots the shipped `cordis.yml`, stores `agent-presets.default`, and asserts an unnamed session composes from it — the package suite proves the layering against a hand-built context, this proves the roster and the settings provider are wired to each other. That test also pins the settings row at a temp file: it defaulted to `$DSH_HOME/settings.yaml`, so a developer's own stored default decided the outcome of a file whose whole point is that only the shipped root does. The Agent Note records the per-resolution read and its correspondence with the session header, and the vacuous-test finding above.
2026-08-04 00:07:11 +08:00
// `default` here is the COMPOSITION default — the base layer the settings
// document overrides. No `roots` entry: the plugin bundles the shipped
// presets itself and prepends their root.
{ id: 'agent-presets', config: { default: 'standard', includeUserRoot: false } },
feat(web): author agent presets from a settings page A composition is a file, but "edit it on the filesystem" is not a browser affordance. The roster gains `read`/`write`/`remove` beside `select`, and the browser gains a settings section over them: the presets as rows, one composition open in a YAML editor at a time, and per-row default, duplicate, and delete. All four authoring methods are loopback-pinned. A composition names the plugins a session runs, so reading one is reconnaissance, writing one is arbitrary capability, and selecting one can move a session onto a preset that edits the live runtime. `agentPreset.list` deliberately stays ordinary and now reports `authorable`, so a surface knows whether creating is possible at all rather than offering a button whose save always fails. Authoring starts by duplicating: a shipped preset opens read-only because the deployment's copy is what a broken local one is compared against. Ids are contained before they become directory names, and the text is parsed with the loader's own schema, so a save cannot leave a file no session could load. Fixes a defect the real-composition test found: a preset written under the user's home could never mount, because the loader resolves a row against the composition's own directory and Node's `node_modules` walk from there never reaches the installed harness. The mount now records the host base and sends bare specifiers there, leaving relative paths resolving from the preset. Also closes the coverage the earlier surfaces in this stack shipped without — the General row, the composer seat, and the plugin halves now have tests.
2026-08-04 12:23:40 +08:00
...extra,
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
]
// The surface is patch layers over an empty preset root, so the root sits
// outside this workspace and bare plugin names cannot resolve by Node's
// upward walk. The flat fallback the preset boot maintains is what makes
// them resolvable — the same mechanism, not a test-only shim.
const home = dirname(settingsFile)
await healProfilesModuleFallback(INSTALL_ANCHOR, home)
const profileDir = join(home, 'profiles', 'spec')
await mkdir(profileDir, { recursive: true })
// Product Bundles are installed into the Profile, not the dsh app. Model
// pnpm's package link for only the selected products; their own production
// dependencies resolve from the linked workspace packages, while shared
// peers still resolve through the installation fallback above.
for (const packageDir of profilePackages) {
const manifest = JSON.parse(await readFile(join(packageDir, 'package.json'), 'utf8')) as { name: string }
const link = join(profileDir, 'node_modules', manifest.name)
await mkdir(dirname(link), { recursive: true })
await symlink(packageDir, link, 'junction')
}
let bundlePatches: PatchOptions[] = [
...loadOverlayPatches('dsh-test', BASE_PATCH),
...loadOverlayPatches('dsh-test', WEB_PATCH),
]
if (profileBundles !== undefined) {
await writeFile(join(profileDir, 'package.json'), JSON.stringify({
private: true,
dependencies: Object.fromEntries(profileBundles.map(name => [name, 'workspace:*'])),
dsh: { profile: { bundles: profileBundles } },
}, null, 2) + '\n')
const profile = loadProfile('dsh-test', 'spec', INSTALL_ANCHOR, home, { userLayer: false })
bundlePatches = profile.layers.flatMap(layer => layer.patches)
}
const rootConfig = join(profileDir, 'cordis.yml')
await writeFile(rootConfig, '[]\n')
return await boot('dsh-test', rootConfig, [...bundlePatches, ...overrides], (bootCtx) => {
provideCmdline(bootCtx, { args: [], exit: () => {} })
})
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
}
const toolNames = (ctx: Context, agent?: Agent): string[] =>
ctx.tools.schemas(agent).map(schema => schema.name).sort()
function toolParameterNames(ctx: Context, agent: Agent, toolName: string): string[] {
const schema = ctx.tools.schemas(agent).find(tool => tool.name === toolName)
if (schema === undefined) throw new Error(`missing tool schema ${toolName}`)
const properties = schema.parameters.properties
if (typeof properties !== 'object' || properties === null || Array.isArray(properties)) {
throw new Error(`${toolName} has invalid parameter properties`)
}
return Object.keys(properties).sort()
}
function enablePresetTool(composition: string, id: string): string {
const row = ` - id: ${id}\n`
const start = composition.indexOf(row)
if (start < 0) throw new Error(`missing preset row ${id}`)
const end = composition.indexOf('\n - id:', start + row.length)
const disabled = composition.indexOf(' disabled: true\n', start)
if (disabled < 0 || (end >= 0 && disabled > end)) {
throw new Error(`preset row ${id} is not disabled`)
}
return composition.slice(0, disabled) + composition.slice(disabled + ' disabled: true\n'.length)
}
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
let ctx: Context
beforeAll(async () => {
feat(agent-presets): make the default preset a user setting `config.default` becomes the composition base of an `agent-presets` settings namespace, so the user document layers over the deployment's engineering default and a person can change which preset new sessions get without a restart. The value is read per resolution rather than snapshotted: a hot-reloaded document takes effect on the next session created, and every running session stays on the preset it was composed from — which is the same rule the session-header guard enforces from the other side. `resolve()` read `config.default` directly, which would have made the whole setting inert; it now goes through `defaultId` like every other caller. The write-protection test is rewritten against a temp profile root. It was passing vacuously: the un-overridden Loader REWRITES the composition it read — stamping `disabled: true` onto the self-disposing row — so the committed fixture had been mutated by the very run that proved the bug, and every later run compared against the damaged file and passed. Building the preset in a temp directory makes the assertion immune to its own failure mode, and it now fails with a visible `+ disabled: true` when the override is removed. Review follow-ups on this layer. The exported schema is `AgentPresetSettingsSchema`, symmetric with the `AgentPresetSettings` interface it resolves and self-describing at an import site. The `session.create` JSDoc promised "the deployment's default preset" for an omitted `agentPreset`, which this layer makes false — it now names the effective default. The constructor records why it does not use `installSettingsSection`: that helper re-judges what a consumer DERIVED across attach and detach, and nothing here is derived. The provider-unload test disposes the fiber `ctx.plugin()` handed back instead of reaching into `ctx.reflect.store`, and the write-protection wait says why slack is the right shape for an absence assertion. The real composition covers the layering too. `apps/cli` boots the shipped `cordis.yml`, stores `agent-presets.default`, and asserts an unnamed session composes from it — the package suite proves the layering against a hand-built context, this proves the roster and the settings provider are wired to each other. That test also pins the settings row at a temp file: it defaulted to `$DSH_HOME/settings.yaml`, so a developer's own stored default decided the outcome of a file whose whole point is that only the shipped root does. The Agent Note records the per-resolution read and its correspondence with the session header, and the vacuous-test finding above.
2026-08-04 00:07:11 +08:00
const settingsFile = join(await mkdtemp(join(tmpdir(), 'dsh-web-presets-')), 'settings.yaml')
await writeFile(settingsFile, '{}\n')
ctx = await bootWeb(settingsFile)
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
}, 120_000)
describe('the shipped Web composition', () => {
it('leaves the global tool layer empty', () => {
// Every model-facing tool belongs to a preset, `ask_user_question`
// included: a tool in the global layer reaches EVERY agent regardless of
// which preset composed it, so a two-tool benchmark surface would really
// present three. A regression here means an agent-plane row came back to
// the host composition.
expect(toolNames(ctx)).toEqual([])
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
})
fix(preset): keep the token meter host-plane and name unjoined agents Moving the agent plane behind presets left two readers on the wrong side of the host/agent line. `dsh-token-meter` was disabled on the host and mounted inside each preset's `compaction` realm, but its three projection units register into the process-wide `sessionProjections` table. A unit registered from one preset answers for every session, so whether a `minimal` session showed a context meter depended on whether some other session had mounted `standard` since boot, and a process that only ever ran `minimal` showed none. The meter takes no configuration, keys every fold by Session, and registers no tool or prompt section, so it returns to the host composition and leaves the presets' `isolate` map; the realm and `compact-basic` stay, because what a preset chooses is whether its agent compacts, not whether its tokens are counted. Nothing named an agent that joined no preset. The join is a scope-parent link, and without it the tools, prompt-section, and skill views resolve the empty global layer: the agent publishes, the turn runs, and the model receives nothing. `AgentPresets` now logs one warning per such agent while a roster is configured, and the invariant companion fails outright — at `system-prompt/assemble` rather than at publication, because an unjoined agent is legal until it addresses a model and `recompose` binds exactly such an agent. The warning stays advisory: a synchronous `agent/created` throw vetoes publication, and the ACP bridge, SDK server, and headless bundle all create an unjoined agent today. Three limits are recorded rather than fixed: projection key presence is not a per-session capability signal, a superseded standing generation is never reclaimed, and a `cordis_mount` temporary plugin belongs to the composition rather than the session that mounted it. Fixes #2203
2026-08-10 22:36:06 +08:00
it('keeps the token meter and its context-meter projections on the host plane', async () => {
// Read before any preset in this file mounts, which is what makes this an
// ownership assertion rather than a mount-order coincidence: a preset-side
// meter sits behind an `isolate` realm and is invisible to `ctx.get`.
//
// The projection registry is process-wide rather than scope-layered, so a
// preset-side meter would also make the browser's context meter appear for
// a `minimal` session the moment some OTHER session mounted a preset that
// carries one, and vanish entirely in a process that only ever ran
// `minimal`. Host ownership is what makes the meter a per-session fact.
expect(ctx.get('tokenMeter')).toBeDefined()
const projections = ctx.get('sessionProjections')
if (projections === undefined) throw new Error('the Web composition must compose a projection registry')
const handle = await ctx.agents.create({
sessionId: SessionId('preset-minimal-meter'),
setup: agentCtx => ctx.agentPresets.mount(agentCtx, 'minimal').then(() => undefined),
})
try {
fix(apiproxy): serve history events and projections from one log position Review found two defects in the previous commit's ordering fix. @pku-xht: `historyStateFor` copied the attached session's events, the handler then awaited `presenterScopeFor`, and only then read the projection baseline off the still-live Session. An append during that await served events cut at N beside a baseline folded to N+1 — one response describing two moments. The same restructure had also moved the baseline read outside the `try`, so a failing snapshot escaped the structured `internal` error. Both awaits now happen before the cut: `historySourceFor` resolves which session serves the read, `presenterScopeFor` ensures the recorded composition, and `historyCutOf` then reads events and baseline adjacently with nothing between them. The whole sequence is back inside the try. The invariant judged any scoped assembly with a chain of one as an unjoined agent, which rejects a legitimate assembly in a standing preset key (that key has no parent of its own). It now gates on `context.agent` — a scope-only read is not an agent and is out of range by construction rather than by a premise about who else calls `assemble` — and asks the roster's own `composedPreset()` instead of introspecting chain length. The advisory warning uses the same relation. Also from review: the `2026-08-05-per-agent-tool-presentation` note still described `presentAs` as per-agent, which standing mounts made false and this branch's own rewording contradicts; the duplicated "process-wide unit table" argument collapses to the Agent Note with pointers from the five copies; a dead `.sort()` before `arrayContaining`; and change-history narration in the tool-cordis README.
2026-08-11 10:45:15 +08:00
// A subset assertion: `tasks`, `goal`, and the rest register into the
// same process-wide table, and this is about the meter's three units.
expect(Object.keys(projections.snapshot(handle.agent.session).values))
fix(preset): keep the token meter host-plane and name unjoined agents Moving the agent plane behind presets left two readers on the wrong side of the host/agent line. `dsh-token-meter` was disabled on the host and mounted inside each preset's `compaction` realm, but its three projection units register into the process-wide `sessionProjections` table. A unit registered from one preset answers for every session, so whether a `minimal` session showed a context meter depended on whether some other session had mounted `standard` since boot, and a process that only ever ran `minimal` showed none. The meter takes no configuration, keys every fold by Session, and registers no tool or prompt section, so it returns to the host composition and leaves the presets' `isolate` map; the realm and `compact-basic` stay, because what a preset chooses is whether its agent compacts, not whether its tokens are counted. Nothing named an agent that joined no preset. The join is a scope-parent link, and without it the tools, prompt-section, and skill views resolve the empty global layer: the agent publishes, the turn runs, and the model receives nothing. `AgentPresets` now logs one warning per such agent while a roster is configured, and the invariant companion fails outright — at `system-prompt/assemble` rather than at publication, because an unjoined agent is legal until it addresses a model and `recompose` binds exactly such an agent. The warning stays advisory: a synchronous `agent/created` throw vetoes publication, and the ACP bridge, SDK server, and headless bundle all create an unjoined agent today. Three limits are recorded rather than fixed: projection key presence is not a per-session capability signal, a superseded standing generation is never reclaimed, and a `cordis_mount` temporary plugin belongs to the composition rather than the session that mounted it. Fixes #2203
2026-08-10 22:36:06 +08:00
.toEqual(expect.arrayContaining(['contextBreakdown', 'contextPressure', 'tokenUsage']))
} finally {
await handle.dispose()
}
})
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
it('supplies both shipped presets, and only those, from the system root', async () => {
const listed = await ctx.agentPresets.list()
expect(listed.map(preset => preset.id).sort()).toEqual(['code', 'cordis', 'minimal', 'standard'])
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
expect(listed.every(preset => preset.trust === 'system')).toBe(true)
expect(ctx.agentPresets.defaultId).toBe('standard')
})
it('composes the full agent from `standard`', async () => {
const handle = await ctx.agents.create({
sessionId: SessionId('preset-standard'),
setup: agentCtx => ctx.agentPresets.mount(agentCtx, 'standard').then(() => undefined),
})
try {
// The EXACT catalog, not a spot-check: an omission is this design's
// quietest failure mode, because a row that registers into the wrong
// layer mounts cleanly and simply contributes nothing. `glob`/`grep` are
// excluded for the reason the TUI composition e2e excludes them — they
// depend on ripgrep being present on the machine.
expect(toolNames(ctx, handle.agent).filter(name => name !== 'glob' && name !== 'grep')).toEqual([
'ask_user_question', 'bash', 'create_goal', 'edit', 'exit_plan_mode',
'get_goal', 'interrupt_agent', 'job_kill', 'job_list', 'job_output', 'list_agents', 'ralph', 'read', 'read_image', 'send_message', 'skill',
'subagent', 'subagent_fork', 'todo_write', 'update_goal', 'web_search',
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
'workflow', 'write',
])
expect(ctx.commands.find(handle.agent, 'goal')).toBeDefined()
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
} finally {
await handle.dispose()
}
})
it('applies the default-off subagent model-selection preference only to new sessions', async () => {
await ctx.settings.update(SUBAGENT_MODEL_SELECTION_SETTINGS_NAMESPACE, { enabled: false })
const disabled = await ctx.agents.create({
sessionId: SessionId('preset-model-selection-disabled'),
setup: agentCtx => ctx.agentPresets.mount(agentCtx, 'standard').then(() => undefined),
})
await ctx.settings.update(SUBAGENT_MODEL_SELECTION_SETTINGS_NAMESPACE, { enabled: true })
const enabled = await ctx.agents.create({
sessionId: SessionId('preset-model-selection-enabled'),
setup: agentCtx => ctx.agentPresets.mount(agentCtx, 'standard').then(() => undefined),
})
try {
expect(toolNames(ctx, disabled.agent)).not.toContain('list_subagent_models')
expect(toolParameterNames(ctx, disabled.agent, 'subagent')).not.toEqual(expect.arrayContaining([
'model', 'provider', 'reasoning_effort',
]))
expect(toolNames(ctx, enabled.agent)).toContain('list_subagent_models')
expect(toolParameterNames(ctx, enabled.agent, 'subagent')).toEqual(expect.arrayContaining([
'model', 'provider', 'reasoning_effort',
]))
expect(toolNames(ctx, disabled.agent)).not.toContain('list_subagent_models')
} finally {
await ctx.settings.update(SUBAGENT_MODEL_SELECTION_SETTINGS_NAMESPACE, { enabled: false })
await enabled.dispose()
await disabled.dispose()
}
})
it('composes the exact RL prompt and two tools from `minimal`', async () => {
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
const handle = await ctx.agents.create({
sessionId: SessionId('preset-minimal'),
setup: agentCtx => ctx.agentPresets.mount(agentCtx, 'minimal').then(() => undefined),
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
})
try {
const assembly = await ctx.systemPrompt.assemble({ scope: handle.agent })
expect(assembly.sections).toEqual([
{ name: 'deployment:persona', text: MINIMAL_PROMPT },
])
expect(assembly.tools.map(tool => tool.name)).toEqual(['bash', 'str_replace_editor'])
expect(assembly.tools.find(tool => tool.name === 'bash')?.description).toBe(MINIMAL_BASH_DESCRIPTION)
expect(JSON.stringify(assembly.tools.find(tool => tool.name === 'str_replace_editor')?.parameters))
.toContain('Absolute path')
expect(ctx.commands.find(handle.agent, 'goal')).toBeUndefined()
expect(ctx.agentPresets.serviceFor(handle.agent, 'compaction')).toBeUndefined()
expect(handle.agent.ctx.get('compaction')).toBeUndefined()
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
} finally {
await handle.dispose()
}
})
it('keeps two differently composed sessions independent', async () => {
const full = await ctx.agents.create({
sessionId: SessionId('preset-both-full'),
setup: agentCtx => ctx.agentPresets.mount(agentCtx, 'standard').then(() => undefined),
})
const minimal = await ctx.agents.create({
sessionId: SessionId('preset-both-minimal'),
setup: agentCtx => ctx.agentPresets.mount(agentCtx, 'minimal').then(() => undefined),
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
})
try {
expect(toolNames(ctx, minimal.agent)).toEqual(['bash', 'str_replace_editor'])
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
expect(toolNames(ctx, full.agent).length).toBeGreaterThan(10)
await minimal.dispose()
// Tearing the minimal session down leaves the full one whole.
expect(toolNames(ctx, full.agent).length).toBeGreaterThan(10)
expect(toolNames(ctx)).toEqual([])
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
} finally {
await full.dispose()
}
})
feat(agent-presets): ship a cordis agent that can author compositions A third built-in preset: the standard coding agent plus the self-referential Cordis toolset, a persona that explains the two-plane split, and a skill teaching composition authoring. It exists so a person can ask an agent to author another agent. The skill ships INSIDE the preset directory rather than in the user's skill root, and the root is derived from the preset's own `baseUrl` — the loader evaluates `!!js` with `with (ctx)`, so a composition can locate itself. A preset is the unit that gets copied and edited, so its documentation should travel with it. The skill leads with the rule that actually bites: a row publishing a service may not sit loose in a preset, whether a row publishes one is not visible from its name (`tool-bash` provides `bashEnv`), and a consumer left outside its provider's isolate group resolves the host registry and then contributes nothing — the quietest failure this design has. Writing the test surfaced a consequence worth stating: an entry-local realm makes the service invisible to the agent's own scope too, not just to the host. Only rows inside that group resolve it, which is precisely what makes `tool-skill` this agent's own rather than a shared one. The test asserts what is actually observable from outside instead of reaching for the isolated service. TRUST: `cordis_mount` evaluates model-written JavaScript against the live runtime, and a composition this agent writes becomes a preset other sessions mount. Both the preset header and the toolset's own documentation say to treat this as shell access. The tools stay opt-in per session — a test pins that they are absent from every other preset.
2026-08-04 00:36:39 +08:00
it('composes the cordis agent with its own toolset', async () => {
const handle = await ctx.agents.create({
sessionId: SessionId('preset-cordis'),
setup: agentCtx => ctx.agentPresets.mount(agentCtx, 'cordis').then(() => undefined),
})
try {
const tools = toolNames(ctx, handle.agent)
// The self-referential toolset is what distinguishes this preset.
2026-08-13 00:40:38 +08:00
expect(tools).toEqual(expect.arrayContaining([
'cordis_inspect_list', 'cordis_inspect_query', 'cordis_inspect_self',
'cordis_define', 'cordis_run', 'cordis_stop', 'cordis_undefine',
]))
feat(agent-presets): ship a cordis agent that can author compositions A third built-in preset: the standard coding agent plus the self-referential Cordis toolset, a persona that explains the two-plane split, and a skill teaching composition authoring. It exists so a person can ask an agent to author another agent. The skill ships INSIDE the preset directory rather than in the user's skill root, and the root is derived from the preset's own `baseUrl` — the loader evaluates `!!js` with `with (ctx)`, so a composition can locate itself. A preset is the unit that gets copied and edited, so its documentation should travel with it. The skill leads with the rule that actually bites: a row publishing a service may not sit loose in a preset, whether a row publishes one is not visible from its name (`tool-bash` provides `bashEnv`), and a consumer left outside its provider's isolate group resolves the host registry and then contributes nothing — the quietest failure this design has. Writing the test surfaced a consequence worth stating: an entry-local realm makes the service invisible to the agent's own scope too, not just to the host. Only rows inside that group resolve it, which is precisely what makes `tool-skill` this agent's own rather than a shared one. The test asserts what is actually observable from outside instead of reaching for the isolated service. TRUST: `cordis_mount` evaluates model-written JavaScript against the live runtime, and a composition this agent writes becomes a preset other sessions mount. Both the preset header and the toolset's own documentation say to treat this as shell access. The tools stay opt-in per session — a test pins that they are absent from every other preset.
2026-08-04 00:36:39 +08:00
// And it keeps the standard agent's own tools rather than replacing them.
expect(tools).toEqual(expect.arrayContaining(['bash', 'read', 'edit', 'skill']))
expect(tools).not.toContain('str_replace_editor')
expect(ctx.commands.find(handle.agent, 'goal')).toBeDefined()
feat(agent-presets): ship a cordis agent that can author compositions A third built-in preset: the standard coding agent plus the self-referential Cordis toolset, a persona that explains the two-plane split, and a skill teaching composition authoring. It exists so a person can ask an agent to author another agent. The skill ships INSIDE the preset directory rather than in the user's skill root, and the root is derived from the preset's own `baseUrl` — the loader evaluates `!!js` with `with (ctx)`, so a composition can locate itself. A preset is the unit that gets copied and edited, so its documentation should travel with it. The skill leads with the rule that actually bites: a row publishing a service may not sit loose in a preset, whether a row publishes one is not visible from its name (`tool-bash` provides `bashEnv`), and a consumer left outside its provider's isolate group resolves the host registry and then contributes nothing — the quietest failure this design has. Writing the test surfaced a consequence worth stating: an entry-local realm makes the service invisible to the agent's own scope too, not just to the host. Only rows inside that group resolve it, which is precisely what makes `tool-skill` this agent's own rather than a shared one. The test asserts what is actually observable from outside instead of reaching for the isolated service. TRUST: `cordis_mount` evaluates model-written JavaScript against the live runtime, and a composition this agent writes becomes a preset other sessions mount. Both the preset header and the toolset's own documentation say to treat this as shell access. The tools stay opt-in per session — a test pins that they are absent from every other preset.
2026-08-04 00:36:39 +08:00
// The preset's own authoring skill registers into ITS layer of the host
// registry: the cordis agent's view carries it, the global view does not.
const scoped = (await ctx.skills.list({ scope: handle.agent })).map(skill => skill.name)
expect(scoped).toContain('editing-cordis-compositions')
expect((await ctx.skills.list()).map(skill => skill.name)).not.toContain('editing-cordis-compositions')
feat(agent-presets): ship a cordis agent that can author compositions A third built-in preset: the standard coding agent plus the self-referential Cordis toolset, a persona that explains the two-plane split, and a skill teaching composition authoring. It exists so a person can ask an agent to author another agent. The skill ships INSIDE the preset directory rather than in the user's skill root, and the root is derived from the preset's own `baseUrl` — the loader evaluates `!!js` with `with (ctx)`, so a composition can locate itself. A preset is the unit that gets copied and edited, so its documentation should travel with it. The skill leads with the rule that actually bites: a row publishing a service may not sit loose in a preset, whether a row publishes one is not visible from its name (`tool-bash` provides `bashEnv`), and a consumer left outside its provider's isolate group resolves the host registry and then contributes nothing — the quietest failure this design has. Writing the test surfaced a consequence worth stating: an entry-local realm makes the service invisible to the agent's own scope too, not just to the host. Only rows inside that group resolve it, which is precisely what makes `tool-skill` this agent's own rather than a shared one. The test asserts what is actually observable from outside instead of reaching for the isolated service. TRUST: `cordis_mount` evaluates model-written JavaScript against the live runtime, and a composition this agent writes becomes a preset other sessions mount. Both the preset header and the toolset's own documentation say to treat this as shell access. The tools stay opt-in per session — a test pins that they are absent from every other preset.
2026-08-04 00:36:39 +08:00
} finally {
await handle.dispose()
}
})
it('presents `code` as Code Mode without disturbing a native session beside it', async () => {
const coded = await ctx.agents.create({
sessionId: SessionId('preset-code'),
setup: agentCtx => ctx.agentPresets.mount(agentCtx, 'code').then(() => undefined),
})
const native = await ctx.agents.create({
sessionId: SessionId('preset-code-native'),
setup: agentCtx => ctx.agentPresets.mount(agentCtx, 'standard').then(() => undefined),
})
try {
// One tool reaches the MODEL: the transport. The registry's catalog for
// this agent is unchanged — a code mode collapses the presentation, not
// the capabilities — so the assembly is what carries the claim.
const assembly = await ctx.systemPrompt.assemble({ scope: coded.agent })
expect(assembly.tools.map(tool => tool.name)).toEqual(['run_code'])
expect(toolNames(ctx, coded.agent)).not.toContain('str_replace_editor')
expect(ctx.commands.find(coded.agent, 'goal')).toBeDefined()
const sdk = assembly.sections.find(section => section.name === 'tools:sdk')?.text ?? ''
expect(sdk).not.toContain('str_replace_editor')
expect(sdk).toContain('web_search')
// The presentation is this agent's alone: the deployment default is
// native, and the session composed from `standard` still sees it.
const nativeAssembly = await ctx.systemPrompt.assemble({ scope: native.agent })
expect(nativeAssembly.tools.map(tool => tool.name)).toContain('bash')
expect(nativeAssembly.tools.map(tool => tool.name)).not.toContain('run_code')
expect(nativeAssembly.sections.some(section => section.name === 'tools:sdk')).toBe(false)
} finally {
await native.dispose()
await coded.dispose()
}
})
feat(agent-presets): ship a cordis agent that can author compositions A third built-in preset: the standard coding agent plus the self-referential Cordis toolset, a persona that explains the two-plane split, and a skill teaching composition authoring. It exists so a person can ask an agent to author another agent. The skill ships INSIDE the preset directory rather than in the user's skill root, and the root is derived from the preset's own `baseUrl` — the loader evaluates `!!js` with `with (ctx)`, so a composition can locate itself. A preset is the unit that gets copied and edited, so its documentation should travel with it. The skill leads with the rule that actually bites: a row publishing a service may not sit loose in a preset, whether a row publishes one is not visible from its name (`tool-bash` provides `bashEnv`), and a consumer left outside its provider's isolate group resolves the host registry and then contributes nothing — the quietest failure this design has. Writing the test surfaced a consequence worth stating: an entry-local realm makes the service invisible to the agent's own scope too, not just to the host. Only rows inside that group resolve it, which is precisely what makes `tool-skill` this agent's own rather than a shared one. The test asserts what is actually observable from outside instead of reaching for the isolated service. TRUST: `cordis_mount` evaluates model-written JavaScript against the live runtime, and a composition this agent writes becomes a preset other sessions mount. Both the preset header and the toolset's own documentation say to treat this as shell access. The tools stay opt-in per session — a test pins that they are absent from every other preset.
2026-08-04 00:36:39 +08:00
it('keeps the self-referential toolset out of every other preset', async () => {
const handle = await ctx.agents.create({
sessionId: SessionId('preset-no-cordis'),
setup: agentCtx => ctx.agentPresets.mount(agentCtx, 'standard').then(() => undefined),
})
try {
// Editing the live runtime is opt-in per session, not ambient.
2026-08-13 00:40:38 +08:00
expect(toolNames(ctx, handle.agent)).not.toContain('cordis_define')
feat(agent-presets): ship a cordis agent that can author compositions A third built-in preset: the standard coding agent plus the self-referential Cordis toolset, a persona that explains the two-plane split, and a skill teaching composition authoring. It exists so a person can ask an agent to author another agent. The skill ships INSIDE the preset directory rather than in the user's skill root, and the root is derived from the preset's own `baseUrl` — the loader evaluates `!!js` with `with (ctx)`, so a composition can locate itself. A preset is the unit that gets copied and edited, so its documentation should travel with it. The skill leads with the rule that actually bites: a row publishing a service may not sit loose in a preset, whether a row publishes one is not visible from its name (`tool-bash` provides `bashEnv`), and a consumer left outside its provider's isolate group resolves the host registry and then contributes nothing — the quietest failure this design has. Writing the test surfaced a consequence worth stating: an entry-local realm makes the service invisible to the agent's own scope too, not just to the host. Only rows inside that group resolve it, which is precisely what makes `tool-skill` this agent's own rather than a shared one. The test asserts what is actually observable from outside instead of reaching for the isolated service. TRUST: `cordis_mount` evaluates model-written JavaScript against the live runtime, and a composition this agent writes becomes a preset other sessions mount. Both the preset header and the toolset's own documentation say to treat this as shell access. The tools stay opt-in per session — a test pins that they are absent from every other preset.
2026-08-04 00:36:39 +08:00
} finally {
await handle.dispose()
}
})
it('ships the composition-authoring skill inside the preset directory', async () => {
// The preset's skill root is derived from its own `baseUrl`, so the skill
// travels with the directory wherever the preset is installed.
const skill = join(
SHIPPED_PRESET_ROOT, 'cordis', 'skills', 'editing-cordis-compositions', 'SKILL.md',
feat(agent-presets): ship a cordis agent that can author compositions A third built-in preset: the standard coding agent plus the self-referential Cordis toolset, a persona that explains the two-plane split, and a skill teaching composition authoring. It exists so a person can ask an agent to author another agent. The skill ships INSIDE the preset directory rather than in the user's skill root, and the root is derived from the preset's own `baseUrl` — the loader evaluates `!!js` with `with (ctx)`, so a composition can locate itself. A preset is the unit that gets copied and edited, so its documentation should travel with it. The skill leads with the rule that actually bites: a row publishing a service may not sit loose in a preset, whether a row publishes one is not visible from its name (`tool-bash` provides `bashEnv`), and a consumer left outside its provider's isolate group resolves the host registry and then contributes nothing — the quietest failure this design has. Writing the test surfaced a consequence worth stating: an entry-local realm makes the service invisible to the agent's own scope too, not just to the host. Only rows inside that group resolve it, which is precisely what makes `tool-skill` this agent's own rather than a shared one. The test asserts what is actually observable from outside instead of reaching for the isolated service. TRUST: `cordis_mount` evaluates model-written JavaScript against the live runtime, and a composition this agent writes becomes a preset other sessions mount. Both the preset header and the toolset's own documentation say to treat this as shell access. The tools stay opt-in per session — a test pins that they are absent from every other preset.
2026-08-04 00:36:39 +08:00
)
expect((await readFile(skill, 'utf8')).startsWith('---\nname: editing-cordis-compositions')).toBe(true)
})
it('merges the global skill layer into a preset agent\'s catalog, keeping local discovery preset-side', async () => {
const proj = await mkdtemp(join(tmpdir(), 'dsh-preset-skill-proj-'))
await mkdir(join(proj, '.dsh', 'skills', 'project-proof'), { recursive: true })
await writeFile(join(proj, '.dsh', 'skills', 'project-proof', 'SKILL.md'), [
'---',
'name: project-proof',
'description: Proves the preset layer discovers project skills beside global ones.',
'---',
'',
'Project proof body.',
'',
].join('\n'))
const handle = await ctx.agents.create({
// Unique per run: the composition persists into the ambient DSH home,
// and a fixed id would collide with a log an earlier run left there.
sessionId: SessionId(`preset-skills-standard-${randomUUID()}`),
setup: agentCtx => ctx.agentPresets.mount(agentCtx, 'standard').then(() => undefined),
})
try {
// The host (global) view carries the deployment-level provider alone:
// local discovery moved behind the presets with `skill-filesystem`.
expect((await ctx.skills.list({ cwd: proj })).map(skill => skill.name)).toEqual(['dsh-badge'])
// The standard agent's view merges the global layer with its preset's
// own local discovery over the session cwd.
const scoped = (await ctx.skills.list({ cwd: proj, scope: handle.agent })).map(skill => skill.name)
expect(scoped).toContain('dsh-badge')
expect(scoped).toContain('project-proof')
// The preset's own loader tool resolves the global-layer skill.
const loaded = await ctx.tools.execute({
callId: CallId('preset-skills-load'),
name: 'skill',
arguments: { name: 'dsh-badge' },
signal: new AbortController().signal,
agent: handle.agent,
})
expect(loaded.isError).toBe(false)
expect(JSON.stringify(loaded.content)).toContain('powered by dsh')
} finally {
await handle.dispose()
}
})
it('shows a minimal agent the global layer but no loader tool', async () => {
const handle = await ctx.agents.create({
sessionId: SessionId(`preset-skills-minimal-${randomUUID()}`),
setup: agentCtx => ctx.agentPresets.mount(agentCtx, 'minimal').then(() => undefined),
})
try {
// Layer visibility is the registry's; whether an agent can USE skills
// stays the preset's choice — minimal mounts no `tool-skill`, so its
// tool table has no loader even though the global layer is readable.
expect((await ctx.skills.list({ scope: handle.agent })).map(skill => skill.name)).toContain('dsh-badge')
expect(toolNames(ctx, handle.agent)).toEqual(['bash', 'str_replace_editor'])
} finally {
await handle.dispose()
}
})
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
it('never rewrites the preset file it composed from', async () => {
// The Loader persists a tree whose plugin self-disposed, and tearing an
// agent down disposes its whole subtree. Inherited, that rewrote the
// shipped composition — truncating it to `[]` the first time a session
// ended — so `PresetTree` refuses to write at all.
const path = join(SHIPPED_PRESET_ROOT, 'standard', 'agent.cordis.yml')
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
const before = await readFile(path, 'utf8')
const handle = await ctx.agents.create({
sessionId: SessionId('preset-readonly'),
setup: agentCtx => ctx.agentPresets.mount(agentCtx, 'standard').then(() => undefined),
})
await handle.dispose()
fix(web): correct the preset-layer contracts review found stale None of these change behavior; each said something that was not true. `SessionCwdConflict`'s doc block had been left stranded above the `AgentPresetConflict` inserted under it, so one class carried a comment about the other and the second carried none. The roster comment named a `.system` directory that does not exist; the shipped root is `config/agent-presets/`, and `system` is the trust its entries carry. The real-composition test attributed the disabled `api-gateway` row to "side effects outside this process" alongside the port and the exporter. It is disabled for a different reason — the api-proxy cannot mount in this layer at all — and hiding that behind the same phrase would leave a later layer unable to tell whether the line can come out. One test claimed to refuse an adoption while asserting only that the header records the preset; it now says what it checks. `PERSONA_SECTION`/`PERSONA_ORDER` existed twice, once in the registry that declares the slot and once restated in the row that replaces it — a drift that would land a preset's persona beside the deployment's instead of shadowing it. The registry exports them now. The preset conflict message read "already runs agent preset undefined" for a session that records none, which is the shape a deployment with no roster produces; it names that case instead, with the regression that reaches it through the gateway. Finally, `PresetTree.write()` drops the `loader/config-update` the inherited method emits — recorded on the override, since a future edit-while-running flow needs its own persistence path.
2026-08-06 13:43:32 +08:00
// Slack, not a race the number has to win. The write is driven by the
// Loader's fiber-unload listener, which fires as the subtree's fibers
// settle rather than when `dispose()` resolves, and the Loader exposes no
// flush to await. A regression writes synchronously inside that listener,
// so any wait past settlement fails; a longer one only slows the test.
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
await new Promise(resolve => setTimeout(resolve, 50))
expect(await readFile(path, 'utf8')).toBe(before)
})
})
describe('product Bundle and user-preset intersection', () => {
const presetIds = ['products-none', 'products-codex', 'products-claude', 'products-both'] as const
type Product = 'codex' | 'claude-code'
type PresetId = typeof presetIds[number]
async function bootProducts(installed: readonly Product[]): Promise<Context> {
const root = await mkdtemp(join(tmpdir(), 'dsh-product-presets-'))
const userRoot = join(root, 'presets')
const settingsFile = join(root, 'settings.yaml')
const standard = await readFile(join(SHIPPED_PRESET_ROOT, 'standard', 'agent.cordis.yml'), 'utf8')
await writeFile(settingsFile, '{}\n')
for (const id of presetIds) {
let composition = standard
if (id === 'products-codex' || id === 'products-both') {
composition = enablePresetTool(composition, 'tool-subagent-codex')
}
if (id === 'products-claude' || id === 'products-both') {
composition = enablePresetTool(composition, 'tool-subagent-claude-code')
}
const directory = join(userRoot, id)
await mkdir(directory, { recursive: true })
await writeFile(join(directory, 'agent.cordis.yml'), composition)
}
const packageDir = (product: Product): string => (
product === 'codex' ? CODEX_PACKAGE_DIR : CLAUDE_CODE_PACKAGE_DIR
)
const packageName = (product: Product): string => (
product === 'codex'
? '@deepseek-ai/dsh-subagent-codex'
: '@deepseek-ai/dsh-subagent-claude-code'
)
return await bootWeb(settingsFile, [
{
id: 'agent-presets',
config: {
default: 'standard',
// The shipped root is the plugin's own, prepended before this.
roots: [{ path: userRoot, trust: 'user' }],
includeUserRoot: false,
},
},
], installed.map(packageDir), [
'@deepseek-ai/dsh-base',
'@deepseek-ai/dsh-web-app',
...installed.map(packageName),
])
}
it('composes the intersection of installed Bundles and enabled preset rows', async () => {
const enabledByPreset: Record<PresetId, Product[]> = {
'products-none': [],
'products-codex': ['codex'],
'products-claude': ['claude-code'],
'products-both': ['codex', 'claude-code'],
}
const scenarios: Array<{ installed: Product[]; presets: readonly PresetId[] }> = [
{ installed: [], presets: ['products-both'] },
{ installed: ['codex'], presets: ['products-both'] },
{ installed: ['claude-code'], presets: ['products-both'] },
{ installed: ['codex', 'claude-code'], presets: presetIds },
]
for (const { installed, presets } of scenarios) {
const productCtx = await bootProducts(installed)
const spawn = vi.spyOn(productCtx.subprocess, 'spawn')
try {
expect(productCtx.subagents.list()
.filter(name => name === 'codex' || name === 'claude-code')
.sort())
.toEqual([...installed].sort())
for (const id of presets) {
const handle = await productCtx.agents.create({
sessionId: SessionId(`preset-${id}-${installed.join('-') || 'none'}-${randomUUID()}`),
setup: agentCtx => productCtx.agentPresets.mount(agentCtx, id).then(() => undefined),
})
try {
const productTools = enabledByPreset[id]
.filter(product => installed.includes(product))
.map(product => product === 'codex' ? 'subagent_codex' : 'subagent_claude_code')
.sort()
2026-08-14 16:18:47 +08:00
const tools = toolNames(productCtx, handle.agent)
expect(tools.filter(name => name === 'subagent_codex' || name === 'subagent_claude_code'))
.toEqual(productTools)
expect(tools).toEqual(expect.arrayContaining(['job_kill', 'job_list', 'job_output']))
for (const productTool of productTools) {
expect(toolParameterNames(productCtx, handle.agent, productTool)).toEqual([
'description', 'prompt', 'run_in_background',
])
}
} finally {
await handle.dispose()
}
}
expect(spawn).not.toHaveBeenCalled()
} finally {
spawn.mockRestore()
await productCtx.fiber.dispose()
}
}
}, 120_000)
it('applies a product-row edit only to later sessions on the preset', async () => {
const productCtx = await bootProducts(['codex'])
const preset = await productCtx.agentPresets.resolve('products-none')
const original = await readFile(preset.path, 'utf8')
const existing = await productCtx.agents.create({
sessionId: SessionId('preset-product-generation-existing'),
setup: agentCtx => productCtx.agentPresets.mount(agentCtx, 'products-none').then(() => undefined),
})
try {
expect(toolNames(productCtx, existing.agent)).not.toContain('subagent_codex')
await writeFile(preset.path, enablePresetTool(original, 'tool-subagent-codex'))
const later = await productCtx.agents.create({
sessionId: SessionId('preset-product-generation-later'),
setup: agentCtx => productCtx.agentPresets.mount(agentCtx, 'products-none').then(() => undefined),
})
try {
expect(toolNames(productCtx, existing.agent)).not.toContain('subagent_codex')
expect(toolNames(productCtx, later.agent)).toContain('subagent_codex')
} finally {
await later.dispose()
}
} finally {
await existing.dispose()
await writeFile(preset.path, original)
await productCtx.fiber.dispose()
}
}, 120_000)
})
describe('a switch survives the session', () => {
it('records the choice so the log states what the agent runs', async () => {
const handle = await ctx.agents.create({
sessionId: SessionId('preset-switch-logged'),
meta: { agentPreset: 'standard' },
setup: agentCtx => ctx.agentPresets.mount(agentCtx, 'standard').then(() => undefined),
})
try {
// The api-proxy's select does exactly this pair while the session is blank.
expect(ctx.commands.find(handle.agent, 'goal')).toBeDefined()
await ctx.agentPresets.recompose(handle.agent.ctx, 'minimal')
handle.agent.session.append('agent-preset/selected', { agentPreset: 'minimal' })
expect(ctx.commands.find(handle.agent, 'goal')).toBeUndefined()
// The header keeps the creation fact; the log carries what it runs.
expect(handle.agent.session.header.agentPreset).toBe('standard')
expect(ctx.sessionProjections.stateOf(handle.agent.session, 'agentPreset')).toBe('minimal')
} finally {
await handle.dispose()
}
})
})
describe('a forked session', () => {
it('inherits the composition its seeded history was produced under', async () => {
const parent = await ctx.agents.create({
sessionId: SessionId('preset-fork-parent'),
meta: { agentPreset: 'minimal' },
setup: agentCtx => ctx.agentPresets.mount(agentCtx, 'minimal').then(() => undefined),
})
const inherited = ctx.sessionProjections.stateOf(parent.agent.session, 'agentPreset') ?? undefined
const child = await ctx.agents.create({
sessionId: SessionId('preset-fork-child'),
meta: {
parentSession: SessionId('preset-fork-parent'),
seedLength: 0,
...inherited === undefined ? {} : { agentPreset: inherited },
},
setup: agentCtx => ctx.agentPresets.mount(agentCtx, inherited).then(() => undefined),
})
try {
// Composing nothing would leave the child empty: this layer moved every
// model-facing row out of the host plane, so there is nothing to inherit
// for free any more.
expect(toolNames(ctx, child.agent)).toEqual(toolNames(ctx, parent.agent))
expect(toolNames(ctx, child.agent).length).toBeGreaterThan(0)
} finally {
await child.dispose()
await parent.dispose()
}
})
})
fix(subagent): compose children from their parent's preset Tool and prompt-section visibility is inherited along dsh-scope's parent chain, and an agent's scope key is minted with no parent. Per-session agent presets moved every model-facing row onto the agent plane and made AgentPresets.mount() the one thing that binds that link, from the api-proxy's session create, resume, and fork paths. The two in-process subagent drivers installed only the per-child persona and tool filter, so a child's scope chain had length one and its registry view resolved the global layer alone — which is empty wherever a preset roster is composed. One-shot children reached the model with no tools, continuable ones with only the host-plane `report`, and neither carried its parent's persona, workspace context, or skill catalog. AgentPresets.composeFrom() joins one agent to the standing composition another already runs on. It is a bind, not a mount: the child gets its parent's exact generation, so a composition edited since the parent started cannot fork it onto another one, and it is synchronous, which is what lets a child creation window use it. applyChildComposition() now takes the parent and performs the join first, making a child composed without it unrepresentable at the call sites. childSessionMeta() records the joined id so a cold read rebuilds the composition the child actually ran under. The audit that followed found two api-proxy readers on the wrong authority: presenterScopeFor() and the live-agent branch of assertPresetUnchanged() both read header.agentPreset, which goes stale the moment a blank session switches preset. A switched session's cold transcript resolved presenters in the older composition's layer and silently degraded to generic cards, and the gateway refused to adopt a live session under the preset it actually runs while accepting the one it left. Both now resolve through resolveSessionPreset(), matching the resume branch fifteen lines above. The owning architecture Agent Note carried the stale claim that the header records what a session runs; it is corrected to name the header/log pair and its three readers. Fixes #2165
2026-08-10 17:46:34 +08:00
describe('a delegated child', () => {
it('runs on the composition its parent runs on', async () => {
const parent = await ctx.agents.create({
sessionId: SessionId('preset-child-parent'),
meta: { agentPreset: 'standard' },
setup: agentCtx => ctx.agentPresets.mount(agentCtx, 'standard').then(() => undefined),
})
// Exactly what an in-process subagent driver's creation window does.
const child = await parent.agent.ctx.agents.create({
sessionId: SessionId('preset-child'),
meta: childSessionMeta(parent.agent, 1, 0),
setup: (agentCtx) => {
applyChildComposition(agentCtx, parent.agent, {})
},
})
try {
expect(toolNames(ctx, child.agent)).toEqual(toolNames(ctx, parent.agent))
// The shipped `standard` preset is the whole coding agent; an empty
// child here is the defect, and equality alone would not catch it.
expect(toolNames(ctx, child.agent)).toContain('bash')
expect(child.agent.session.header.agentPreset).toBe('standard')
} finally {
await child.dispose()
await parent.dispose()
}
})
it('follows a parent that switched preset while blank', async () => {
const parent = await ctx.agents.create({
sessionId: SessionId('preset-child-switch-parent'),
meta: { agentPreset: 'standard' },
setup: agentCtx => ctx.agentPresets.mount(agentCtx, 'standard').then(() => undefined),
})
await ctx.agentPresets.recompose(parent.agent.ctx, 'minimal')
const child = await parent.agent.ctx.agents.create({
sessionId: SessionId('preset-child-switch'),
meta: childSessionMeta(parent.agent, 1, 0),
setup: (agentCtx) => {
applyChildComposition(agentCtx, parent.agent, {})
},
})
try {
// The live scope chain is the authority, not the parent's creation
// header — which still names `standard`.
expect(toolNames(ctx, child.agent)).toEqual(toolNames(ctx, parent.agent))
expect(child.agent.session.header.agentPreset).toBe('minimal')
} finally {
await child.dispose()
await parent.dispose()
}
})
})
describe('a launcher that configures no writable root', () => {
// The claim this default exists for, asserted through the real shipped
// bundles rather than a hand-built context: `apps/cli` patches in only the
// system root, and a person's own presets are found anyway because the
// roster derives `<dshHome>/.agent-presets` itself. `$DSH_HOME` is pointed
// at a temp home BEFORE boot — the derived root is resolved when the plugin
// is constructed, and an unpinned run would read the developer's own.
let derivedCtx: Context
let previousHome: string | undefined
beforeAll(async () => {
const home = await mkdtemp(join(tmpdir(), 'dsh-preset-derived-'))
previousHome = process.env.DSH_HOME
process.env.DSH_HOME = home
await mkdir(join(home, '.agent-presets', 'derived-mine'), { recursive: true })
await writeFile(
join(home, '.agent-presets', 'derived-mine', 'agent.cordis.yml'),
'- id: tool-todo\n name: \'@deepseek-ai/dsh-tool-todo\'\n config:\n allowParallelInProgress: true\n',
)
const settingsFile = join(await mkdtemp(join(tmpdir(), 'dsh-preset-derived-settings-')), 'settings.yaml')
await writeFile(settingsFile, '{}\n')
// No configured roots: the shipped one is the plugin's own, and the
// writable one is the roster's own default rather than this patch's job.
derivedCtx = await bootWeb(settingsFile, [{
id: 'agent-presets',
config: { default: 'standard', includeUserRoot: true },
}])
}, 120_000)
afterAll(async () => {
if (previousHome === undefined) delete process.env.DSH_HOME
else process.env.DSH_HOME = previousHome
await derivedCtx.fiber.dispose()
})
it('discovers and mounts a preset the person authored under the harness home', async () => {
const listed = await derivedCtx.agentPresets.list()
const mine = listed.find(preset => preset.id === 'derived-mine')
expect(mine).toMatchObject({ trust: 'user' })
// Omitted rather than undefined: a healthy row carries no `broken` key.
expect(mine?.broken).toBeUndefined()
expect(derivedCtx.agentPresets.authorable).toBe(true)
const handle = await derivedCtx.agents.create({
sessionId: SessionId('preset-derived-root'),
setup: agentCtx => derivedCtx.agentPresets.mount(agentCtx, 'derived-mine').then(() => undefined),
})
try {
expect(toolNames(derivedCtx, handle.agent)).toContain('todo_write')
} finally {
await handle.dispose()
}
})
})
feat(web): author agent presets from a settings page A composition is a file, but "edit it on the filesystem" is not a browser affordance. The roster gains `read`/`write`/`remove` beside `select`, and the browser gains a settings section over them: the presets as rows, one composition open in a YAML editor at a time, and per-row default, duplicate, and delete. All four authoring methods are loopback-pinned. A composition names the plugins a session runs, so reading one is reconnaissance, writing one is arbitrary capability, and selecting one can move a session onto a preset that edits the live runtime. `agentPreset.list` deliberately stays ordinary and now reports `authorable`, so a surface knows whether creating is possible at all rather than offering a button whose save always fails. Authoring starts by duplicating: a shipped preset opens read-only because the deployment's copy is what a broken local one is compared against. Ids are contained before they become directory names, and the text is parsed with the loader's own schema, so a save cannot leave a file no session could load. Fixes a defect the real-composition test found: a preset written under the user's home could never mount, because the loader resolves a row against the composition's own directory and Node's `node_modules` walk from there never reaches the installed harness. The mount now records the host base and sends bare specifiers there, leaving relative paths resolving from the preset. Also closes the coverage the earlier surfaces in this stack shipped without — the General row, the composer seat, and the plugin halves now have tests.
2026-08-04 12:23:40 +08:00
describe('authoring a preset on the shipped composition', () => {
let authorCtx: Context
let userRoot: string
beforeAll(async () => {
userRoot = join(await mkdtemp(join(tmpdir(), 'dsh-preset-authoring-')), 'profiles')
feat(web): author agent presets from a settings page A composition is a file, but "edit it on the filesystem" is not a browser affordance. The roster gains `read`/`write`/`remove` beside `select`, and the browser gains a settings section over them: the presets as rows, one composition open in a YAML editor at a time, and per-row default, duplicate, and delete. All four authoring methods are loopback-pinned. A composition names the plugins a session runs, so reading one is reconnaissance, writing one is arbitrary capability, and selecting one can move a session onto a preset that edits the live runtime. `agentPreset.list` deliberately stays ordinary and now reports `authorable`, so a surface knows whether creating is possible at all rather than offering a button whose save always fails. Authoring starts by duplicating: a shipped preset opens read-only because the deployment's copy is what a broken local one is compared against. Ids are contained before they become directory names, and the text is parsed with the loader's own schema, so a save cannot leave a file no session could load. Fixes a defect the real-composition test found: a preset written under the user's home could never mount, because the loader resolves a row against the composition's own directory and Node's `node_modules` walk from there never reaches the installed harness. The mount now records the host base and sends bare specifiers there, leaving relative paths resolving from the preset. Also closes the coverage the earlier surfaces in this stack shipped without — the General row, the composer seat, and the plugin halves now have tests.
2026-08-04 12:23:40 +08:00
const settingsFile = join(await mkdtemp(join(tmpdir(), 'dsh-preset-authoring-settings-')), 'settings.yaml')
await writeFile(settingsFile, '{}\n')
authorCtx = await bootWeb(settingsFile, [{
id: 'agent-presets',
config: {
default: 'standard',
// The root does not exist yet: a deployment whose user has authored
// nothing is the normal first-run state. The shipped root is the
// plugin's own, prepended before this.
roots: [{ path: userRoot, trust: 'user' }],
includeUserRoot: false,
feat(web): author agent presets from a settings page A composition is a file, but "edit it on the filesystem" is not a browser affordance. The roster gains `read`/`write`/`remove` beside `select`, and the browser gains a settings section over them: the presets as rows, one composition open in a YAML editor at a time, and per-row default, duplicate, and delete. All four authoring methods are loopback-pinned. A composition names the plugins a session runs, so reading one is reconnaissance, writing one is arbitrary capability, and selecting one can move a session onto a preset that edits the live runtime. `agentPreset.list` deliberately stays ordinary and now reports `authorable`, so a surface knows whether creating is possible at all rather than offering a button whose save always fails. Authoring starts by duplicating: a shipped preset opens read-only because the deployment's copy is what a broken local one is compared against. Ids are contained before they become directory names, and the text is parsed with the loader's own schema, so a save cannot leave a file no session could load. Fixes a defect the real-composition test found: a preset written under the user's home could never mount, because the loader resolves a row against the composition's own directory and Node's `node_modules` walk from there never reaches the installed harness. The mount now records the host base and sends bare specifiers there, leaving relative paths resolving from the preset. Also closes the coverage the earlier surfaces in this stack shipped without — the General row, the composer seat, and the plugin halves now have tests.
2026-08-04 12:23:40 +08:00
},
}])
})
refactor(agent-presets,web): copy-only preset authoring with a path to the files The web YAML editor is gone. agentPreset.write (arbitrary composition text) became agentPreset.copy { from, agentPreset, name? }: a host-side whole-directory copy of ids the host resolves itself — symlinks dereferenced, modes re-tightened to owner-only with owner-execute kept, metadata rewritten to keep the source's description but never its name or roster order. No composition text or path crosses the wire in either authoring direction, and the entryListSchema/!!js concern dissolves with assertComposition itself. The settings section becomes: a read-only viewer over shipped compositions, a copy dialog (id + optional display name) as the only create entry, delete for custom rows, and a location action leading into the preset's own files — agentPreset.openDocument { agentPreset } resolves the directory host-side and opens it natively, or answers { opened: false, path } for the row to show as text where the deployment has no desktop. agentPreset.list reports hasDocument beside authorable; the gateway's nativeOpen config pins the capability where canOpenNativePath platform detection would mislead. The privileged set is now read/copy/openDocument/remove. With files as the only composition editor, standing mounts grew stamp-keyed generations: ensureStanding compares the composition file's mtime+size and starts the next generation for later sessions, while every joined session keeps the generation it runs on. New keyless web lane (agent-preset-authoring, overlay pins nativeOpen: false so goldens render one branch on every platform) drives view/copy/reveal/delete end to end; the real-composition CLI e2e switches to copy semantics.
2026-08-08 22:35:26 +08:00
it('refuses to copy over or delete a shipped preset', async () => {
await expect(authorCtx.agentPresets.copy('minimal', 'standard')).rejects.toThrow(/already exists/)
feat(web): author agent presets from a settings page A composition is a file, but "edit it on the filesystem" is not a browser affordance. The roster gains `read`/`write`/`remove` beside `select`, and the browser gains a settings section over them: the presets as rows, one composition open in a YAML editor at a time, and per-row default, duplicate, and delete. All four authoring methods are loopback-pinned. A composition names the plugins a session runs, so reading one is reconnaissance, writing one is arbitrary capability, and selecting one can move a session onto a preset that edits the live runtime. `agentPreset.list` deliberately stays ordinary and now reports `authorable`, so a surface knows whether creating is possible at all rather than offering a button whose save always fails. Authoring starts by duplicating: a shipped preset opens read-only because the deployment's copy is what a broken local one is compared against. Ids are contained before they become directory names, and the text is parsed with the loader's own schema, so a save cannot leave a file no session could load. Fixes a defect the real-composition test found: a preset written under the user's home could never mount, because the loader resolves a row against the composition's own directory and Node's `node_modules` walk from there never reaches the installed harness. The mount now records the host base and sends bare specifiers there, leaving relative paths resolving from the preset. Also closes the coverage the earlier surfaces in this stack shipped without — the General row, the composer seat, and the plugin halves now have tests.
2026-08-04 12:23:40 +08:00
await expect(authorCtx.agentPresets.remove('standard')).rejects.toThrow(/ships with the deployment/)
})
it.each(['../escape', 'a/b', '/abs', 'Upper'])('refuses the uncontainable id %j', async (id) => {
// The id becomes a directory name under the user root, so containment is
// checked on the id rather than on the joined path afterwards.
refactor(agent-presets,web): copy-only preset authoring with a path to the files The web YAML editor is gone. agentPreset.write (arbitrary composition text) became agentPreset.copy { from, agentPreset, name? }: a host-side whole-directory copy of ids the host resolves itself — symlinks dereferenced, modes re-tightened to owner-only with owner-execute kept, metadata rewritten to keep the source's description but never its name or roster order. No composition text or path crosses the wire in either authoring direction, and the entryListSchema/!!js concern dissolves with assertComposition itself. The settings section becomes: a read-only viewer over shipped compositions, a copy dialog (id + optional display name) as the only create entry, delete for custom rows, and a location action leading into the preset's own files — agentPreset.openDocument { agentPreset } resolves the directory host-side and opens it natively, or answers { opened: false, path } for the row to show as text where the deployment has no desktop. agentPreset.list reports hasDocument beside authorable; the gateway's nativeOpen config pins the capability where canOpenNativePath platform detection would mislead. The privileged set is now read/copy/openDocument/remove. With files as the only composition editor, standing mounts grew stamp-keyed generations: ensureStanding compares the composition file's mtime+size and starts the next generation for later sessions, while every joined session keeps the generation it runs on. New keyless web lane (agent-preset-authoring, overlay pins nativeOpen: false so goldens render one branch on every platform) drives view/copy/reveal/delete end to end; the real-composition CLI e2e switches to copy semantics.
2026-08-08 22:35:26 +08:00
await expect(authorCtx.agentPresets.copy('minimal', id)).rejects.toThrow()
feat(web): author agent presets from a settings page A composition is a file, but "edit it on the filesystem" is not a browser affordance. The roster gains `read`/`write`/`remove` beside `select`, and the browser gains a settings section over them: the presets as rows, one composition open in a YAML editor at a time, and per-row default, duplicate, and delete. All four authoring methods are loopback-pinned. A composition names the plugins a session runs, so reading one is reconnaissance, writing one is arbitrary capability, and selecting one can move a session onto a preset that edits the live runtime. `agentPreset.list` deliberately stays ordinary and now reports `authorable`, so a surface knows whether creating is possible at all rather than offering a button whose save always fails. Authoring starts by duplicating: a shipped preset opens read-only because the deployment's copy is what a broken local one is compared against. Ids are contained before they become directory names, and the text is parsed with the loader's own schema, so a save cannot leave a file no session could load. Fixes a defect the real-composition test found: a preset written under the user's home could never mount, because the loader resolves a row against the composition's own directory and Node's `node_modules` walk from there never reaches the installed harness. The mount now records the host base and sends bare specifiers there, leaving relative paths resolving from the preset. Also closes the coverage the earlier surfaces in this stack shipped without — the General row, the composer seat, and the plugin halves now have tests.
2026-08-04 12:23:40 +08:00
})
refactor(agent-presets,web): copy-only preset authoring with a path to the files The web YAML editor is gone. agentPreset.write (arbitrary composition text) became agentPreset.copy { from, agentPreset, name? }: a host-side whole-directory copy of ids the host resolves itself — symlinks dereferenced, modes re-tightened to owner-only with owner-execute kept, metadata rewritten to keep the source's description but never its name or roster order. No composition text or path crosses the wire in either authoring direction, and the entryListSchema/!!js concern dissolves with assertComposition itself. The settings section becomes: a read-only viewer over shipped compositions, a copy dialog (id + optional display name) as the only create entry, delete for custom rows, and a location action leading into the preset's own files — agentPreset.openDocument { agentPreset } resolves the directory host-side and opens it natively, or answers { opened: false, path } for the row to show as text where the deployment has no desktop. agentPreset.list reports hasDocument beside authorable; the gateway's nativeOpen config pins the capability where canOpenNativePath platform detection would mislead. The privileged set is now read/copy/openDocument/remove. With files as the only composition editor, standing mounts grew stamp-keyed generations: ensureStanding compares the composition file's mtime+size and starts the next generation for later sessions, while every joined session keeps the generation it runs on. New keyless web lane (agent-preset-authoring, overlay pins nativeOpen: false so goldens render one branch on every platform) drives view/copy/reveal/delete end to end; the real-composition CLI e2e switches to copy semantics.
2026-08-08 22:35:26 +08:00
it('copies a shipped preset a session then really composes from', async () => {
await authorCtx.agentPresets.copy('minimal', 'my-agent', '我的模式')
feat(web): author agent presets from a settings page A composition is a file, but "edit it on the filesystem" is not a browser affordance. The roster gains `read`/`write`/`remove` beside `select`, and the browser gains a settings section over them: the presets as rows, one composition open in a YAML editor at a time, and per-row default, duplicate, and delete. All four authoring methods are loopback-pinned. A composition names the plugins a session runs, so reading one is reconnaissance, writing one is arbitrary capability, and selecting one can move a session onto a preset that edits the live runtime. `agentPreset.list` deliberately stays ordinary and now reports `authorable`, so a surface knows whether creating is possible at all rather than offering a button whose save always fails. Authoring starts by duplicating: a shipped preset opens read-only because the deployment's copy is what a broken local one is compared against. Ids are contained before they become directory names, and the text is parsed with the loader's own schema, so a save cannot leave a file no session could load. Fixes a defect the real-composition test found: a preset written under the user's home could never mount, because the loader resolves a row against the composition's own directory and Node's `node_modules` walk from there never reaches the installed harness. The mount now records the host base and sends bare specifiers there, leaving relative paths resolving from the preset. Also closes the coverage the earlier surfaces in this stack shipped without — the General row, the composer seat, and the plugin halves now have tests.
2026-08-04 12:23:40 +08:00
refactor(agent-presets,web): copy-only preset authoring with a path to the files The web YAML editor is gone. agentPreset.write (arbitrary composition text) became agentPreset.copy { from, agentPreset, name? }: a host-side whole-directory copy of ids the host resolves itself — symlinks dereferenced, modes re-tightened to owner-only with owner-execute kept, metadata rewritten to keep the source's description but never its name or roster order. No composition text or path crosses the wire in either authoring direction, and the entryListSchema/!!js concern dissolves with assertComposition itself. The settings section becomes: a read-only viewer over shipped compositions, a copy dialog (id + optional display name) as the only create entry, delete for custom rows, and a location action leading into the preset's own files — agentPreset.openDocument { agentPreset } resolves the directory host-side and opens it natively, or answers { opened: false, path } for the row to show as text where the deployment has no desktop. agentPreset.list reports hasDocument beside authorable; the gateway's nativeOpen config pins the capability where canOpenNativePath platform detection would mislead. The privileged set is now read/copy/openDocument/remove. With files as the only composition editor, standing mounts grew stamp-keyed generations: ensureStanding compares the composition file's mtime+size and starts the next generation for later sessions, while every joined session keeps the generation it runs on. New keyless web lane (agent-preset-authoring, overlay pins nativeOpen: false so goldens render one branch on every platform) drives view/copy/reveal/delete end to end; the real-composition CLI e2e switches to copy semantics.
2026-08-08 22:35:26 +08:00
// Round-trips through the roster as a `user` row carrying the given name
// and the source's description, over the source's own composition text.
feat(web): author agent presets from a settings page A composition is a file, but "edit it on the filesystem" is not a browser affordance. The roster gains `read`/`write`/`remove` beside `select`, and the browser gains a settings section over them: the presets as rows, one composition open in a YAML editor at a time, and per-row default, duplicate, and delete. All four authoring methods are loopback-pinned. A composition names the plugins a session runs, so reading one is reconnaissance, writing one is arbitrary capability, and selecting one can move a session onto a preset that edits the live runtime. `agentPreset.list` deliberately stays ordinary and now reports `authorable`, so a surface knows whether creating is possible at all rather than offering a button whose save always fails. Authoring starts by duplicating: a shipped preset opens read-only because the deployment's copy is what a broken local one is compared against. Ids are contained before they become directory names, and the text is parsed with the loader's own schema, so a save cannot leave a file no session could load. Fixes a defect the real-composition test found: a preset written under the user's home could never mount, because the loader resolves a row against the composition's own directory and Node's `node_modules` walk from there never reaches the installed harness. The mount now records the host base and sends bare specifiers there, leaving relative paths resolving from the preset. Also closes the coverage the earlier surfaces in this stack shipped without — the General row, the composer seat, and the plugin halves now have tests.
2026-08-04 12:23:40 +08:00
const preset = await authorCtx.agentPresets.resolve('my-agent')
refactor(agent-presets,web): copy-only preset authoring with a path to the files The web YAML editor is gone. agentPreset.write (arbitrary composition text) became agentPreset.copy { from, agentPreset, name? }: a host-side whole-directory copy of ids the host resolves itself — symlinks dereferenced, modes re-tightened to owner-only with owner-execute kept, metadata rewritten to keep the source's description but never its name or roster order. No composition text or path crosses the wire in either authoring direction, and the entryListSchema/!!js concern dissolves with assertComposition itself. The settings section becomes: a read-only viewer over shipped compositions, a copy dialog (id + optional display name) as the only create entry, delete for custom rows, and a location action leading into the preset's own files — agentPreset.openDocument { agentPreset } resolves the directory host-side and opens it natively, or answers { opened: false, path } for the row to show as text where the deployment has no desktop. agentPreset.list reports hasDocument beside authorable; the gateway's nativeOpen config pins the capability where canOpenNativePath platform detection would mislead. The privileged set is now read/copy/openDocument/remove. With files as the only composition editor, standing mounts grew stamp-keyed generations: ensureStanding compares the composition file's mtime+size and starts the next generation for later sessions, while every joined session keeps the generation it runs on. New keyless web lane (agent-preset-authoring, overlay pins nativeOpen: false so goldens render one branch on every platform) drives view/copy/reveal/delete end to end; the real-composition CLI e2e switches to copy semantics.
2026-08-08 22:35:26 +08:00
const source = await authorCtx.agentPresets.resolve('minimal')
feat(web): author agent presets from a settings page A composition is a file, but "edit it on the filesystem" is not a browser affordance. The roster gains `read`/`write`/`remove` beside `select`, and the browser gains a settings section over them: the presets as rows, one composition open in a YAML editor at a time, and per-row default, duplicate, and delete. All four authoring methods are loopback-pinned. A composition names the plugins a session runs, so reading one is reconnaissance, writing one is arbitrary capability, and selecting one can move a session onto a preset that edits the live runtime. `agentPreset.list` deliberately stays ordinary and now reports `authorable`, so a surface knows whether creating is possible at all rather than offering a button whose save always fails. Authoring starts by duplicating: a shipped preset opens read-only because the deployment's copy is what a broken local one is compared against. Ids are contained before they become directory names, and the text is parsed with the loader's own schema, so a save cannot leave a file no session could load. Fixes a defect the real-composition test found: a preset written under the user's home could never mount, because the loader resolves a row against the composition's own directory and Node's `node_modules` walk from there never reaches the installed harness. The mount now records the host base and sends bare specifiers there, leaving relative paths resolving from the preset. Also closes the coverage the earlier surfaces in this stack shipped without — the General row, the composer seat, and the plugin halves now have tests.
2026-08-04 12:23:40 +08:00
expect(preset.trust).toBe('user')
refactor(agent-presets,web): copy-only preset authoring with a path to the files The web YAML editor is gone. agentPreset.write (arbitrary composition text) became agentPreset.copy { from, agentPreset, name? }: a host-side whole-directory copy of ids the host resolves itself — symlinks dereferenced, modes re-tightened to owner-only with owner-execute kept, metadata rewritten to keep the source's description but never its name or roster order. No composition text or path crosses the wire in either authoring direction, and the entryListSchema/!!js concern dissolves with assertComposition itself. The settings section becomes: a read-only viewer over shipped compositions, a copy dialog (id + optional display name) as the only create entry, delete for custom rows, and a location action leading into the preset's own files — agentPreset.openDocument { agentPreset } resolves the directory host-side and opens it natively, or answers { opened: false, path } for the row to show as text where the deployment has no desktop. agentPreset.list reports hasDocument beside authorable; the gateway's nativeOpen config pins the capability where canOpenNativePath platform detection would mislead. The privileged set is now read/copy/openDocument/remove. With files as the only composition editor, standing mounts grew stamp-keyed generations: ensureStanding compares the composition file's mtime+size and starts the next generation for later sessions, while every joined session keeps the generation it runs on. New keyless web lane (agent-preset-authoring, overlay pins nativeOpen: false so goldens render one branch on every platform) drives view/copy/reveal/delete end to end; the real-composition CLI e2e switches to copy semantics.
2026-08-08 22:35:26 +08:00
expect(preset.name).toBe('我的模式')
expect(preset.description).toBe(source.description)
expect(await authorCtx.agentPresets.read('my-agent')).toBe(await authorCtx.agentPresets.read('minimal'))
feat(web): author agent presets from a settings page A composition is a file, but "edit it on the filesystem" is not a browser affordance. The roster gains `read`/`write`/`remove` beside `select`, and the browser gains a settings section over them: the presets as rows, one composition open in a YAML editor at a time, and per-row default, duplicate, and delete. All four authoring methods are loopback-pinned. A composition names the plugins a session runs, so reading one is reconnaissance, writing one is arbitrary capability, and selecting one can move a session onto a preset that edits the live runtime. `agentPreset.list` deliberately stays ordinary and now reports `authorable`, so a surface knows whether creating is possible at all rather than offering a button whose save always fails. Authoring starts by duplicating: a shipped preset opens read-only because the deployment's copy is what a broken local one is compared against. Ids are contained before they become directory names, and the text is parsed with the loader's own schema, so a save cannot leave a file no session could load. Fixes a defect the real-composition test found: a preset written under the user's home could never mount, because the loader resolves a row against the composition's own directory and Node's `node_modules` walk from there never reaches the installed harness. The mount now records the host base and sends bare specifiers there, leaving relative paths resolving from the preset. Also closes the coverage the earlier surfaces in this stack shipped without — the General row, the composer seat, and the plugin halves now have tests.
2026-08-04 12:23:40 +08:00
// Owner-only, in an owner-only directory: a composition is executable
// configuration on a machine that may have other users.
expect((await stat(preset.path)).mode & 0o777).toBe(0o600)
const handle = await authorCtx.agents.create({
sessionId: SessionId('preset-authored'),
setup: agentCtx => authorCtx.agentPresets.mount(agentCtx, 'my-agent').then(() => undefined),
})
try {
refactor(agent-presets,web): copy-only preset authoring with a path to the files The web YAML editor is gone. agentPreset.write (arbitrary composition text) became agentPreset.copy { from, agentPreset, name? }: a host-side whole-directory copy of ids the host resolves itself — symlinks dereferenced, modes re-tightened to owner-only with owner-execute kept, metadata rewritten to keep the source's description but never its name or roster order. No composition text or path crosses the wire in either authoring direction, and the entryListSchema/!!js concern dissolves with assertComposition itself. The settings section becomes: a read-only viewer over shipped compositions, a copy dialog (id + optional display name) as the only create entry, delete for custom rows, and a location action leading into the preset's own files — agentPreset.openDocument { agentPreset } resolves the directory host-side and opens it natively, or answers { opened: false, path } for the row to show as text where the deployment has no desktop. agentPreset.list reports hasDocument beside authorable; the gateway's nativeOpen config pins the capability where canOpenNativePath platform detection would mislead. The privileged set is now read/copy/openDocument/remove. With files as the only composition editor, standing mounts grew stamp-keyed generations: ensureStanding compares the composition file's mtime+size and starts the next generation for later sessions, while every joined session keeps the generation it runs on. New keyless web lane (agent-preset-authoring, overlay pins nativeOpen: false so goldens render one branch on every platform) drives view/copy/reveal/delete end to end; the real-composition CLI e2e switches to copy semantics.
2026-08-08 22:35:26 +08:00
// The same tools the shipped `minimal` composes, from a directory copied
feat(web): author agent presets from a settings page A composition is a file, but "edit it on the filesystem" is not a browser affordance. The roster gains `read`/`write`/`remove` beside `select`, and the browser gains a settings section over them: the presets as rows, one composition open in a YAML editor at a time, and per-row default, duplicate, and delete. All four authoring methods are loopback-pinned. A composition names the plugins a session runs, so reading one is reconnaissance, writing one is arbitrary capability, and selecting one can move a session onto a preset that edits the live runtime. `agentPreset.list` deliberately stays ordinary and now reports `authorable`, so a surface knows whether creating is possible at all rather than offering a button whose save always fails. Authoring starts by duplicating: a shipped preset opens read-only because the deployment's copy is what a broken local one is compared against. Ids are contained before they become directory names, and the text is parsed with the loader's own schema, so a save cannot leave a file no session could load. Fixes a defect the real-composition test found: a preset written under the user's home could never mount, because the loader resolves a row against the composition's own directory and Node's `node_modules` walk from there never reaches the installed harness. The mount now records the host base and sends bare specifiers there, leaving relative paths resolving from the preset. Also closes the coverage the earlier surfaces in this stack shipped without — the General row, the composer seat, and the plugin halves now have tests.
2026-08-04 12:23:40 +08:00
// through the service into a root outside the installed harness.
expect(toolNames(authorCtx, handle.agent)).toEqual(['bash', 'str_replace_editor'])
feat(web): author agent presets from a settings page A composition is a file, but "edit it on the filesystem" is not a browser affordance. The roster gains `read`/`write`/`remove` beside `select`, and the browser gains a settings section over them: the presets as rows, one composition open in a YAML editor at a time, and per-row default, duplicate, and delete. All four authoring methods are loopback-pinned. A composition names the plugins a session runs, so reading one is reconnaissance, writing one is arbitrary capability, and selecting one can move a session onto a preset that edits the live runtime. `agentPreset.list` deliberately stays ordinary and now reports `authorable`, so a surface knows whether creating is possible at all rather than offering a button whose save always fails. Authoring starts by duplicating: a shipped preset opens read-only because the deployment's copy is what a broken local one is compared against. Ids are contained before they become directory names, and the text is parsed with the loader's own schema, so a save cannot leave a file no session could load. Fixes a defect the real-composition test found: a preset written under the user's home could never mount, because the loader resolves a row against the composition's own directory and Node's `node_modules` walk from there never reaches the installed harness. The mount now records the host base and sends bare specifiers there, leaving relative paths resolving from the preset. Also closes the coverage the earlier surfaces in this stack shipped without — the General row, the composer seat, and the plugin halves now have tests.
2026-08-04 12:23:40 +08:00
} finally {
await handle.dispose()
}
})
refactor(agent-presets,web): copy-only preset authoring with a path to the files The web YAML editor is gone. agentPreset.write (arbitrary composition text) became agentPreset.copy { from, agentPreset, name? }: a host-side whole-directory copy of ids the host resolves itself — symlinks dereferenced, modes re-tightened to owner-only with owner-execute kept, metadata rewritten to keep the source's description but never its name or roster order. No composition text or path crosses the wire in either authoring direction, and the entryListSchema/!!js concern dissolves with assertComposition itself. The settings section becomes: a read-only viewer over shipped compositions, a copy dialog (id + optional display name) as the only create entry, delete for custom rows, and a location action leading into the preset's own files — agentPreset.openDocument { agentPreset } resolves the directory host-side and opens it natively, or answers { opened: false, path } for the row to show as text where the deployment has no desktop. agentPreset.list reports hasDocument beside authorable; the gateway's nativeOpen config pins the capability where canOpenNativePath platform detection would mislead. The privileged set is now read/copy/openDocument/remove. With files as the only composition editor, standing mounts grew stamp-keyed generations: ensureStanding compares the composition file's mtime+size and starts the next generation for later sessions, while every joined session keeps the generation it runs on. New keyless web lane (agent-preset-authoring, overlay pins nativeOpen: false so goldens render one branch on every platform) drives view/copy/reveal/delete end to end; the real-composition CLI e2e switches to copy semantics.
2026-08-08 22:35:26 +08:00
it('deletes what it copied', async () => {
await authorCtx.agentPresets.copy('minimal', 'doomed')
feat(web): author agent presets from a settings page A composition is a file, but "edit it on the filesystem" is not a browser affordance. The roster gains `read`/`write`/`remove` beside `select`, and the browser gains a settings section over them: the presets as rows, one composition open in a YAML editor at a time, and per-row default, duplicate, and delete. All four authoring methods are loopback-pinned. A composition names the plugins a session runs, so reading one is reconnaissance, writing one is arbitrary capability, and selecting one can move a session onto a preset that edits the live runtime. `agentPreset.list` deliberately stays ordinary and now reports `authorable`, so a surface knows whether creating is possible at all rather than offering a button whose save always fails. Authoring starts by duplicating: a shipped preset opens read-only because the deployment's copy is what a broken local one is compared against. Ids are contained before they become directory names, and the text is parsed with the loader's own schema, so a save cannot leave a file no session could load. Fixes a defect the real-composition test found: a preset written under the user's home could never mount, because the loader resolves a row against the composition's own directory and Node's `node_modules` walk from there never reaches the installed harness. The mount now records the host base and sends bare specifiers there, leaving relative paths resolving from the preset. Also closes the coverage the earlier surfaces in this stack shipped without — the General row, the composer seat, and the plugin halves now have tests.
2026-08-04 12:23:40 +08:00
await authorCtx.agentPresets.remove('doomed')
expect((await authorCtx.agentPresets.list()).map(preset => preset.id)).not.toContain('doomed')
})
})
feat(agent-presets): make the default preset a user setting `config.default` becomes the composition base of an `agent-presets` settings namespace, so the user document layers over the deployment's engineering default and a person can change which preset new sessions get without a restart. The value is read per resolution rather than snapshotted: a hot-reloaded document takes effect on the next session created, and every running session stays on the preset it was composed from — which is the same rule the session-header guard enforces from the other side. `resolve()` read `config.default` directly, which would have made the whole setting inert; it now goes through `defaultId` like every other caller. The write-protection test is rewritten against a temp profile root. It was passing vacuously: the un-overridden Loader REWRITES the composition it read — stamping `disabled: true` onto the self-disposing row — so the committed fixture had been mutated by the very run that proved the bug, and every later run compared against the damaged file and passed. Building the preset in a temp directory makes the assertion immune to its own failure mode, and it now fails with a visible `+ disabled: true` when the override is removed. Review follow-ups on this layer. The exported schema is `AgentPresetSettingsSchema`, symmetric with the `AgentPresetSettings` interface it resolves and self-describing at an import site. The `session.create` JSDoc promised "the deployment's default preset" for an omitted `agentPreset`, which this layer makes false — it now names the effective default. The constructor records why it does not use `installSettingsSection`: that helper re-judges what a consumer DERIVED across attach and detach, and nothing here is derived. The provider-unload test disposes the fiber `ctx.plugin()` handed back instead of reaching into `ctx.reflect.store`, and the write-protection wait says why slack is the right shape for an absence assertion. The real composition covers the layering too. `apps/cli` boots the shipped `cordis.yml`, stores `agent-presets.default`, and asserts an unnamed session composes from it — the package suite proves the layering against a hand-built context, this proves the roster and the settings provider are wired to each other. That test also pins the settings row at a temp file: it defaulted to `$DSH_HOME/settings.yaml`, so a developer's own stored default decided the outcome of a file whose whole point is that only the shipped root does. The Agent Note records the per-resolution read and its correspondence with the session header, and the vacuous-test finding above.
2026-08-04 00:07:11 +08:00
/**
* Which preset an unnamed session gets is a user setting layered over the
* composition's own default. The package suite proves the layering against a
* hand-built context; this proves it through the shipped `cordis.yml` — that
* the roster and the settings provider are actually wired to each other, and
* that the id the setting names is the one a session composes from.
*/
describe('the default preset as a user setting', () => {
it('composes an unnamed session from the stored default, not the composed one', async () => {
expect(ctx.agentPresets.defaultId).toBe('standard')
await ctx.settings.update(settingsNamespace(SETTINGS_NAMESPACE), { default: 'minimal' })
feat(agent-presets): make the default preset a user setting `config.default` becomes the composition base of an `agent-presets` settings namespace, so the user document layers over the deployment's engineering default and a person can change which preset new sessions get without a restart. The value is read per resolution rather than snapshotted: a hot-reloaded document takes effect on the next session created, and every running session stays on the preset it was composed from — which is the same rule the session-header guard enforces from the other side. `resolve()` read `config.default` directly, which would have made the whole setting inert; it now goes through `defaultId` like every other caller. The write-protection test is rewritten against a temp profile root. It was passing vacuously: the un-overridden Loader REWRITES the composition it read — stamping `disabled: true` onto the self-disposing row — so the committed fixture had been mutated by the very run that proved the bug, and every later run compared against the damaged file and passed. Building the preset in a temp directory makes the assertion immune to its own failure mode, and it now fails with a visible `+ disabled: true` when the override is removed. Review follow-ups on this layer. The exported schema is `AgentPresetSettingsSchema`, symmetric with the `AgentPresetSettings` interface it resolves and self-describing at an import site. The `session.create` JSDoc promised "the deployment's default preset" for an omitted `agentPreset`, which this layer makes false — it now names the effective default. The constructor records why it does not use `installSettingsSection`: that helper re-judges what a consumer DERIVED across attach and detach, and nothing here is derived. The provider-unload test disposes the fiber `ctx.plugin()` handed back instead of reaching into `ctx.reflect.store`, and the write-protection wait says why slack is the right shape for an absence assertion. The real composition covers the layering too. `apps/cli` boots the shipped `cordis.yml`, stores `agent-presets.default`, and asserts an unnamed session composes from it — the package suite proves the layering against a hand-built context, this proves the roster and the settings provider are wired to each other. That test also pins the settings row at a temp file: it defaulted to `$DSH_HOME/settings.yaml`, so a developer's own stored default decided the outcome of a file whose whole point is that only the shipped root does. The Agent Note records the per-resolution read and its correspondence with the session header, and the vacuous-test finding above.
2026-08-04 00:07:11 +08:00
try {
expect(ctx.agentPresets.defaultId).toBe('minimal')
feat(agent-presets): make the default preset a user setting `config.default` becomes the composition base of an `agent-presets` settings namespace, so the user document layers over the deployment's engineering default and a person can change which preset new sessions get without a restart. The value is read per resolution rather than snapshotted: a hot-reloaded document takes effect on the next session created, and every running session stays on the preset it was composed from — which is the same rule the session-header guard enforces from the other side. `resolve()` read `config.default` directly, which would have made the whole setting inert; it now goes through `defaultId` like every other caller. The write-protection test is rewritten against a temp profile root. It was passing vacuously: the un-overridden Loader REWRITES the composition it read — stamping `disabled: true` onto the self-disposing row — so the committed fixture had been mutated by the very run that proved the bug, and every later run compared against the damaged file and passed. Building the preset in a temp directory makes the assertion immune to its own failure mode, and it now fails with a visible `+ disabled: true` when the override is removed. Review follow-ups on this layer. The exported schema is `AgentPresetSettingsSchema`, symmetric with the `AgentPresetSettings` interface it resolves and self-describing at an import site. The `session.create` JSDoc promised "the deployment's default preset" for an omitted `agentPreset`, which this layer makes false — it now names the effective default. The constructor records why it does not use `installSettingsSection`: that helper re-judges what a consumer DERIVED across attach and detach, and nothing here is derived. The provider-unload test disposes the fiber `ctx.plugin()` handed back instead of reaching into `ctx.reflect.store`, and the write-protection wait says why slack is the right shape for an absence assertion. The real composition covers the layering too. `apps/cli` boots the shipped `cordis.yml`, stores `agent-presets.default`, and asserts an unnamed session composes from it — the package suite proves the layering against a hand-built context, this proves the roster and the settings provider are wired to each other. That test also pins the settings row at a temp file: it defaulted to `$DSH_HOME/settings.yaml`, so a developer's own stored default decided the outcome of a file whose whole point is that only the shipped root does. The Agent Note records the per-resolution read and its correspondence with the session header, and the vacuous-test finding above.
2026-08-04 00:07:11 +08:00
const handle = await ctx.agents.create({
sessionId: SessionId('preset-user-default'),
setup: agentCtx => ctx.agentPresets.mount(agentCtx).then(() => undefined),
})
try {
// `mount()` with no id resolves the effective default. Two tools, not
// `standard`'s catalog: the setting decided the composition.
expect(toolNames(ctx, handle.agent)).toEqual(['bash', 'str_replace_editor'])
feat(agent-presets): make the default preset a user setting `config.default` becomes the composition base of an `agent-presets` settings namespace, so the user document layers over the deployment's engineering default and a person can change which preset new sessions get without a restart. The value is read per resolution rather than snapshotted: a hot-reloaded document takes effect on the next session created, and every running session stays on the preset it was composed from — which is the same rule the session-header guard enforces from the other side. `resolve()` read `config.default` directly, which would have made the whole setting inert; it now goes through `defaultId` like every other caller. The write-protection test is rewritten against a temp profile root. It was passing vacuously: the un-overridden Loader REWRITES the composition it read — stamping `disabled: true` onto the self-disposing row — so the committed fixture had been mutated by the very run that proved the bug, and every later run compared against the damaged file and passed. Building the preset in a temp directory makes the assertion immune to its own failure mode, and it now fails with a visible `+ disabled: true` when the override is removed. Review follow-ups on this layer. The exported schema is `AgentPresetSettingsSchema`, symmetric with the `AgentPresetSettings` interface it resolves and self-describing at an import site. The `session.create` JSDoc promised "the deployment's default preset" for an omitted `agentPreset`, which this layer makes false — it now names the effective default. The constructor records why it does not use `installSettingsSection`: that helper re-judges what a consumer DERIVED across attach and detach, and nothing here is derived. The provider-unload test disposes the fiber `ctx.plugin()` handed back instead of reaching into `ctx.reflect.store`, and the write-protection wait says why slack is the right shape for an absence assertion. The real composition covers the layering too. `apps/cli` boots the shipped `cordis.yml`, stores `agent-presets.default`, and asserts an unnamed session composes from it — the package suite proves the layering against a hand-built context, this proves the roster and the settings provider are wired to each other. That test also pins the settings row at a temp file: it defaulted to `$DSH_HOME/settings.yaml`, so a developer's own stored default decided the outcome of a file whose whole point is that only the shipped root does. The Agent Note records the per-resolution read and its correspondence with the session header, and the vacuous-test finding above.
2026-08-04 00:07:11 +08:00
} finally {
await handle.dispose()
}
} finally {
// The context is shared with the rest of the file. `replace({})` drops
// the user section wholesale so the field re-inherits the composition
// base; `update` merges, and would leave the override standing.
await ctx.settings.replace(settingsNamespace(SETTINGS_NAMESPACE), {})
}
expect(ctx.agentPresets.defaultId).toBe('standard')
})
})
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
describe('a session keeps the preset it was created with', () => {
it('refuses to adopt a live session under a different preset', async () => {
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
const handle = await ctx.agents.create({
sessionId: SessionId('preset-locked'),
meta: { agentPreset: 'minimal' },
setup: agentCtx => ctx.agentPresets.mount(agentCtx, 'minimal').then(() => undefined),
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
})
try {
// The api-proxy guard reads exactly this: the header records what the
// session runs, so naming anything else is a caller error rather than a
// switch. Its history was produced under `minimal`'s two tools.
expect(handle.agent.session.header.agentPreset).toBe('minimal')
feat(web): move the agent plane behind per-session presets The Web overlay disables base's 32 agent-plane rows and mounts the preset roster instead, so each session composes its own tools and prompt rather than sharing one process-wide set. The TUI keeps base unchanged: it is single-session and composing its agent process-wide is correct there. `roots` is patched in by AppCLIEntry, like `distIndex`: the shipped presets sit beside the composition that names them and the user's live under the Harness home, neither of which a config author chooses. A session's preset is fixed at creation. Naming a different one for an existing identity is `agent-preset-conflict` rather than a switch, because that session's history was produced under the first preset's tools. The guard sits after `await creation`, beside the cwd check, so it covers every path that yields a live agent — freshly created, adopted live, resumed, or recovered by the concurrent-creation catch. A request naming no preset adopts the session as it is, keeping reconnect and retry ordinary. Two bugs the real-composition test caught, both invisible to unit tests: `PresetTree` now refuses to write. The Loader persists a tree whose plugin self-disposed, and tearing an agent down disposes its whole subtree — inherited, that rewrote the shipped composition, truncating a 241-line preset to `[]` the first time a session ended. `dsh-tool-skill` compared against a lookup of its own name in the global layer, so it threw inside any preset: `register()` files into the calling context's scope. It now compares against the definition it registered, which is what the identity check meant all along. The `standard` catalog is asserted exactly, not spot-checked: a row that registers into the wrong layer mounts cleanly and simply contributes nothing, so an omission is this design's quietest failure. It matches the shipped TUI catalog plus `glob`/`grep`, the pair that composition documents as ripgrep-dependent. Re-records `cordis-inspect-jsdoc`, whose rendered `SessionHeader` gains the `agentPreset` field. `fs-glob-sampling` fails identically on pristine master and is untouched here. The browser e2e scaffold gains the roster fact AppCLIEntry supplies. `roots` is resolved and patched in by the CLI entry, like `distIndex` on the webserver row, and this lane boots the shipped tree without that entry — so it has to supply the same fact or the roster resolves nothing and every session in the lane composes an agent with no tools, no persona, and no token meter. Only the shipped root: a developer's own `~/.dsh/.agent-presets` must not decide a golden. The `cordis:group` builtin comes with it, exactly as `boot()` registers it, because a preset resolving package names from its own directory cannot reach `@cordisjs/plugin-group` by name. The lane stays red through this layer and the next four for the reason stated above — the api-proxy injects `subagents`, `workspace`, and `tools`, so `api-gateway` cannot activate and the browser has no `/api` at all. It goes green again in the layer that returns those registries to the host plane; this change is what makes that layer's fix sufficient rather than partial.
2026-08-03 23:44:09 +08:00
} finally {
await handle.dispose()
}
})
})
describe('a composition that configures its own preset roots', () => {
let rootsCtx: Context
let teamRoot: string
beforeAll(async () => {
const home = await mkdtemp(join(tmpdir(), 'dsh-preset-roots-'))
const settingsFile = join(home, 'settings.yaml')
await writeFile(settingsFile, '{}\n')
// A workspace-shared root beside the deployment: one preset of its own,
// plus a directory that claims a shipped id.
teamRoot = join(home, 'team-presets')
const minimalComposition = await readFile(join(SHIPPED_PRESET_ROOT, 'minimal', 'agent.cordis.yml'), 'utf8')
for (const id of ['team-spec', 'minimal']) {
await mkdir(join(teamRoot, id), { recursive: true })
await writeFile(join(teamRoot, id, 'agent.cordis.yml'), minimalComposition)
}
// The user layer of the reported regression: a profile's cordis.patch.yml
// configuring a shared preset root. The plugin must EXTEND it with its
// own shipped root, never lose it.
rootsCtx = await bootWeb(settingsFile, [{
id: 'agent-presets',
config: {
default: 'standard',
roots: [{ path: teamRoot, trust: 'user' }],
includeUserRoot: false,
},
}])
}, 120_000)
afterAll(async () => {
await rootsCtx.fiber.dispose()
})
it('keeps configured roots alongside the always-prepended shipped root', async () => {
expect(rootsCtx.agentPresets.roots.map(root => root.path)).toEqual([
SHIPPED_PRESET_ROOT,
teamRoot,
])
const listed = await rootsCtx.agentPresets.list()
expect(listed.map(preset => preset.id).sort()).toEqual(['code', 'cordis', 'minimal', 'standard', 'team-spec'])
expect(listed.every(preset => preset.broken === undefined)).toBe(true)
// The shipped root comes first: a configured directory claiming a shipped
// id is shadowed, never the other way around.
expect(listed.find(preset => preset.id === 'minimal')?.trust).toBe('system')
expect(listed.find(preset => preset.id === 'team-spec')?.trust).toBe('user')
})
it('composes an agent from a configured-root preset', async () => {
const handle = await rootsCtx.agents.create({
sessionId: SessionId('preset-team-spec'),
setup: agentCtx => rootsCtx.agentPresets.mount(agentCtx, 'team-spec').then(() => undefined),
})
try {
expect(toolNames(rootsCtx, handle.agent)).toEqual(['bash', 'str_replace_editor'])
} finally {
await handle.dispose()
}
})
})