deepseek-harness/packages/bash/tool-bash/src/index.ts

571 lines
25 KiB
TypeScript
Raw Normal View History

/**
* Model-facing `bash` tool over the `ctx.bash` executor seam. Background calls
* register process handles with `ctx.tasks`; their work uses task cancellation
* rather than the tool-call signal after an id is returned.
*
* TODO(permissions): deployment policy belongs in `tools/pre-execute` and
* sandboxing executors; see docs/architecture.md § Extending The Harness.
* @module @deepseek-ai/dsh-tool-bash
*/
import { Service, type Context } from 'cordis'
import z from 'schemastery'
feat(acp): honor per-session cwd — run each ACP session in its own workspace Lifts the RFC 010 § Deferred restriction that the server had to launch in the workspace ("cwd must equal the launch directory"). An editor can now open any project folder, and N concurrent sessions over one connection can each target a different directory. - packages/acp: drop the `cwd === process.cwd()` guard in validateWorkspaceParams (keep "must be absolute" — the cwd becomes the session header / bash workdir), and drop the persisted-cwd-vs-launch-dir check in session/load (a resumed session keeps its original header.cwd, so its bash tools run in its workspace). - packages/tool-bash: the missing link — default the bash workdir to the calling agent's session cwd (`exec.agent.session.header.cwd`) via a new resolveWorkdir helper. An explicit model `workdir` still wins; a relative one resolves against the session cwd. This is the only correct spot for multi-session: N sessions share one ctx.bash executor, so the workdir must come per-call from exec.agent, not executor config. Falls back to the executor default when no session cwd is available (preserves non-ACP behavior). - Trust: the cwd originates from the ACP client (the user's editor) at session/new — same trust level as the old launch dir; no new untrusted-input path. `additionalDirectories` (scope widening / sandbox) stays rejected. - Tests: bridge accepts any absolute cwd + records it on the header; session/load honors the persisted cwd; bash defaults to / resolves relative against the session cwd; two sessions with different cwds each run bash in their own dir; non-absolute cwd still rejected. 100% per-file coverage maintained. - Docs: RFC 010 status + § Deferred cwd bullet marked RESOLVED; acp README adds a Per-session cwd section; tool-bash + example READMEs and e2e comments updated.
2026-06-17 10:01:18 +08:00
import { isAbsolute, resolve as resolvePath } from 'node:path'
import { defineTool, TOOL_ABORTED } from '@deepseek-ai/dsh-tools'
import type { GenericCallView, TerminalCallView, ToolExecution, ToolResult, ToolResultView } from '@deepseek-ai/dsh-tools'
import { HarnessError } from '@deepseek-ai/dsh-llm'
import type { Agent } from '@deepseek-ai/dsh-agent'
import type {} from '@deepseek-ai/dsh-session-persistence'
import type {} from '@deepseek-ai/dsh-system-prompt'
import type {} from '@deepseek-ai/dsh-tasks'
import type {} from '@deepseek-ai/dsh-user-approval'
import type { SandboxExecutionPolicy, SandboxMode } from '@deepseek-ai/dsh-sandbox'
import { ESCALATION_TARGETS, approveEscalation, canonicalPath, validateEscalationArgs } from '@deepseek-ai/dsh-sandbox'
import type { SandboxPolicyService } from '@deepseek-ai/dsh-sandbox-policy'
Merge remote-tracking branch 'origin/master' into cross-family-fs-sandbox # Conflicts: # .agents/notes/implemented/feature/2026-07-14-cross-family-fs-sandbox.i18n.yaml # .agents/notes/implemented/feature/2026-07-14-cross-family-fs-sandbox.md # .agents/notes/implemented/feature/2026-07-14-cross-family-fs-sandbox.zh.md # docs/capability-seams.md # docs/cordis-catalog/events.md # docs/cordis-catalog/services.md # docs/event-producer-consumer.md # docs/module-graph.md # docs/persistence-catalog.md # docs/rfc/INDEX.md # examples/acp-agent/README.md # examples/acp-agent/fs.cordis.snapshot.yml # examples/acp-agent/fs.cordis.yml # examples/acp-agent/tests/snapshots/escalation-approved/session.jsonl # examples/acp-agent/tests/snapshots/escalation-rejected/session.jsonl # examples/acp-agent/tests/snapshots/hook-cc-pretool-ask/session.jsonl # examples/acp-agent/tests/snapshots/permission-switching/session.jsonl # examples/acp-agent/tests/snapshots/workspace-context/system-prompt.expected.md # examples/acp-agent/tests/snapshots/workspace-context/tool-schemas.expected.json # examples/acp-agent/tests/snapshots/workspace-edit/system-prompt.expected.md # examples/acp-agent/tests/snapshots/workspace-edit/tool-schemas.expected.json # packages/bash/bash/src/index.ts # packages/bash/tool-bash/package.json # packages/bash/tool-bash/src/index.ts # packages/bash/tool-bash/tests/tools.spec.ts # packages/cordis/tool-cordis/src/api-catalog.ts # packages/fs/README.md # packages/fs/tool-fs/src/edit.ts # packages/fs/tool-fs/src/write.ts # packages/sandbox/README.md # pnpm-lock.yaml
2026-07-20 11:40:29 +08:00
import { DSH_ENV_PREFIX } from '@deepseek-ai/dsh-bash'
2026-07-21 03:08:35 +08:00
import type { BashRunResult, DshEnvironment, DshEnvironmentKey } from '@deepseek-ai/dsh-bash'
import { DSH_HOME_ENV, resolveDshHome } from '@deepseek-ai/dsh-paths'
import { processOutcome } from './background.ts'
import { parseExitStatus, renderProcessRead, renderResult } from './render.ts'
declare module 'cordis' {
interface Context {
bashEnv: BashEnvRegistry
}
}
export const name = 'tool-bash'
export const inject = ['tools', 'bash', 'systemPrompt']
/** Configuration for the bash tool and its managed child environment. */
export interface Config {
/** Expose `run_in_background` (default true); disabled calls are also rejected. */
enableRunInBackground?: boolean
/** DeepSeek Harness home directory exposed as `DSH_HOME`; defaults to `$DSH_HOME` or `~/.dsh`. */
dshHome?: string
}
/** Runtime configuration schema for the bash tool plugin. */
export const Config: z<Config> = z.object({
enableRunInBackground: z.boolean().default(true),
dshHome: z.string(),
})
/** Model-visible metadata for one managed `DSH_*` environment variable. */
export interface BashEnvVariable {
/** Concise description of the environment fact represented by the variable. */
description: string
}
/**
* A plugin contribution to the managed environment of each model bash call.
* Declared keys make ownership conflicts detectable before the first command;
* `resolve` computes only the values available for the current execution.
*/
export interface BashEnvContributor {
/** Stable contributor name used in diagnostics and duplicate detection. */
name: string
/** Complete set of `DSH_*` keys this contributor may return. */
variables: Readonly<Record<DshEnvironmentKey, BashEnvVariable>>
/**
* Resolve this contributor's available values for one tool execution.
* @param execution - the bash tool execution and its optional calling agent.
* @returns a partial map containing only keys declared in {@link variables}.
*/
resolve(execution: ToolExecution): Readonly<Partial<Record<DshEnvironmentKey, string>>>
}
/** An enumerable declaration returned by {@link BashEnvRegistry.list}. */
export interface BashEnvVariableInfo extends BashEnvVariable {
/** Contributor that owns the variable. */
contributor: string
/** Declared `DSH_*` environment variable name. */
key: DshEnvironmentKey
}
const DSH_SHELL_KEY = `${DSH_ENV_PREFIX}SHELL` as const
const DSH_SESSION_ID_KEY = `${DSH_ENV_PREFIX}SESSION_ID` as const
const DSH_SESSION_JSONL_KEY = `${DSH_ENV_PREFIX}SESSION_JSONL` as const
const RESERVED_BASH_ENV_KEYS = new Set<DshEnvironmentKey>([
DSH_HOME_ENV,
DSH_SHELL_KEY,
DSH_SESSION_ID_KEY,
])
const BASH_ENV_KEY_SUFFIX = /^[A-Z][A-Z0-9_]*$/
/**
* Registry (`ctx.bashEnv`) for trusted, per-execution `DSH_*` variables.
* The namespace is rebuilt for every model bash call: ambient `DSH_*` values
* are discarded by the executor, then the registry's current snapshot is
* injected. Built-in shell facts remain owned by the registry itself while
* plugins can register additional, enumerable facts with effect-scoped
* disposal.
*/
export class BashEnvRegistry extends Service {
private readonly contributors = new Map<string, BashEnvContributor>()
private readonly keyOwners = new Map<DshEnvironmentKey, string>()
private readonly dshHome: string
/**
* Create and install the `ctx.bashEnv` service.
* @param ctx - Cordis context that owns the service and registrations.
* @param config - home-directory configuration for the built-in variables.
*/
constructor(ctx: Context, config: Config = {}) {
super(ctx, 'bashEnv')
this.dshHome = resolveDshHome(config.dshHome)
}
/**
* Register one environment contributor. Names and keys are unique; built-in
* keys are reserved. Registration is disposed with the calling plugin fiber.
* @param contributor - declared key ownership and per-execution resolver.
* @returns the disposer that unregisters the contribution.
*/
register(contributor: BashEnvContributor): () => void {
const dispose = this.ctx.effect(function* (this: BashEnvRegistry) {
if (contributor.name.trim().length === 0) {
throw new Error('bash env contributor name must be non-empty')
}
if (this.contributors.has(contributor.name)) {
throw new Error(`bash env contributor "${contributor.name}" is already registered`)
}
const variables = Object.entries(contributor.variables) as [DshEnvironmentKey, BashEnvVariable][]
for (const [key, variable] of variables) {
if (!key.startsWith(DSH_ENV_PREFIX)
|| !BASH_ENV_KEY_SUFFIX.test(key.slice(DSH_ENV_PREFIX.length))) {
throw new Error(`bash env contributor "${contributor.name}" declared invalid key "${key}"`)
}
if (RESERVED_BASH_ENV_KEYS.has(key)) {
throw new Error(`bash env contributor "${contributor.name}" cannot own reserved key "${key}"`)
}
if (variable.description.trim().length === 0) {
throw new Error(`bash env contributor "${contributor.name}" must describe "${key}"`)
}
const owner = this.keyOwners.get(key)
if (owner !== undefined) {
throw new Error(`bash env key "${key}" is already owned by contributor "${owner}"; contributor "${contributor.name}" cannot also own it`)
}
}
this.contributors.set(contributor.name, contributor)
for (const [key] of variables) this.keyOwners.set(key, contributor.name)
yield () => {
this.contributors.delete(contributor.name)
for (const [key] of variables) this.keyOwners.delete(key)
}
}.bind(this), 'bashEnv.register()')
return () => void dispose()
}
/**
* Build the trusted `DSH_*` snapshot for one bash tool execution.
* @param execution - the current tool execution.
* @returns an immutable environment overlay containing built-ins and current contributions.
*/
collect(execution: ToolExecution): DshEnvironment {
const values: Record<DshEnvironmentKey, string> = {
[DSH_HOME_ENV]: this.dshHome,
[DSH_SHELL_KEY]: '1',
}
if (execution.agent !== undefined) {
values[DSH_SESSION_ID_KEY] = execution.agent.session.header.id
}
for (const contributor of [...this.contributors.values()].sort((left, right) => left.name.localeCompare(right.name))) {
const resolved = contributor.resolve(execution)
for (const [rawKey, value] of Object.entries(resolved)) {
const key = rawKey as DshEnvironmentKey
if (!Object.hasOwn(contributor.variables, key)) {
throw new Error(`bash env contributor "${contributor.name}" returned undeclared key "${key}"`)
}
if (typeof value !== 'string') {
throw new Error(`bash env contributor "${contributor.name}" returned a non-string value for "${key}"`)
}
values[key] = value
}
}
return Object.freeze(Object.fromEntries(Object.entries(values).sort(([left], [right]) => left.localeCompare(right))))
}
// TODO(bash-env-list-builtins): Include registry-owned built-ins before diagnostics,
// prompt, or UI code treats list() as an exhaustive environment catalog.
/**
* Enumerate plugin-contributed variables without executing their resolvers.
* @returns declarations sorted by environment variable name.
*/
list(): BashEnvVariableInfo[] {
return [...this.contributors.values()]
.flatMap(contributor => Object.entries(contributor.variables).map(([key, variable]) => ({
contributor: contributor.name,
description: variable.description,
key: key as DshEnvironmentKey,
})))
.sort((left, right) => left.key.localeCompare(right.key))
}
}
2026-07-21 01:11:55 +08:00
/** Parsed tool args; execute validates value constraints absent from ParameterSchemaSpec. */
interface BashToolArgs {
command: string
description: string
timeoutMs?: number
workdir?: string
run_in_background?: boolean
sandbox_permissions?: string
justification?: string
}
function validateBashArgs(args: BashToolArgs): void {
if (args.command.trim().length === 0) {
throw new Error('invalid command: expected a non-empty string')
}
if (args.description.trim().length === 0) {
throw new Error('invalid description: expected a non-empty string')
}
if (args.timeoutMs !== undefined && (!Number.isFinite(args.timeoutMs) || args.timeoutMs <= 0)) {
throw new Error(`invalid timeoutMs: expected a positive number, got ${JSON.stringify(args.timeoutMs)}`)
}
feat(sandbox): cross-family file sandbox — one policy home, sandboxed fs provider, fs escalation parity Extend SandboxMode enforcement from bash to the filesystem tools, the sandbox RFC's deferred cross-family phase. - dsh-sandbox-policy (new, ctx.sandboxPolicy): the single home for the deployment default mode + workspaceRoot and the per-session override event, renamed bash/sandbox-mode -> sandbox/mode and moved here with its fold/setter. Decouples the bash seam from dsh-session. - dsh-fs-sandbox (new): SandboxedFileSystem extends LocalFileSystem and fences write/edit by the per-call mode (read-only denies, workspace-write contains to the workspace + temp roots via the shared writableRoots, danger passes through); reads pass through. Structured FS_SANDBOX_DENIED; in-lock parent re-canonicalization. A policy fence in trusted code, not a kernel boundary. - dsh-sandbox: the shared escalation kit (writableRoots, the strictly-wider ladder, denial/hint markers, approveEscalation) both tool families use; approveEscalation takes a structural approver so dsh-sandbox gains no approval/agent dependency, and both tools stay duplication-free. - tool-fs: write/edit advertise sandbox_permissions/justification under a confining ctx.fs, map FS_SANDBOX_DENIED to the shared [sandbox: ...] marker, and resolve the same one-approved-wider retry. - examples/acp-agent: composes sandbox-policy + fs-sandbox, drops the gating that disabled the fs stack under confined modes. RFC docs/rfc/implemented/feature/2026-07-14-cross-family-fs-sandbox.md; the old sandbox RFC's In-process/deferred/FAQ sections updated to shipped fact.
2026-07-14 20:05:57 +08:00
// The escalation pairing (sandbox_permissions ⇔ justification, non-empty) is
// the shared rule both enforcing families validate identically.
validateEscalationArgs(args.sandbox_permissions, args.justification)
}
function bashDescription(backgroundEnabled: boolean, escalationModes: readonly SandboxMode[]): string {
const background = backgroundEnabled
? 'Set `run_in_background: true` for long-running commands: the call returns a task id immediately; read its output with `task_output` and stop it with `task_kill`.'
: 'Background execution is not available; long-running commands must finish within the timeout.'
const base = 'Execute a bash command (`bash -c`) and return its stdout/stderr. '
+ 'Each call runs in a fresh shell: no state (cwd, variables, functions) persists between calls — '
+ 'pass `workdir` instead of using `cd`. Non-zero exits are reported as `[exit code: N]`. '
+ `Current harness environment facts are exposed through managed \`$${DSH_ENV_PREFIX}*\` variables; inspect them when needed. `
+ 'Commands may run under a file sandbox; a blocked file operation is reported as `[sandbox: file access denied under <mode> mode]` — a policy denial, not a bug in the command; do not retry another way. '
+ 'Long output is truncated to its tail; the full output is saved to a file whose path is reported when available. '
+ background
if (escalationModes.length === 0) return base
return base + ' Attempting a command the sandbox may deny is safe and expected: run it and read the '
+ 'marker rather than assuming the denial. When a command is denied and a wider mode would let it '
+ 'succeed, escalate immediately in the same turn — the one sanctioned exception to a denial: retry '
+ 'the exact same command once with `sandbox_permissions` (the narrowest wider mode that suffices) '
+ 'plus a one-sentence `justification`. Do not detour through chat to ask permission first — the '
+ 'approval prompt raised by that retry is how the user consents. If the session states approval '
+ 'prompts are disabled, there is no exception: a denial is final — do not set `sandbox_permissions`. '
+ 'Never escalate speculatively: ground the request in a real denial — normally the one this command '
+ 'just hit; escalating up front is fine only when this session already denied the same access. '
+ 'A rejected escalation is final for that command — stop and explain, never work around '
+ 'it — but it does not forbid attempting or escalating other commands later.'
}
feat(acp): tool-owned tool-call UI presentation (title/command/output) In Zed the tool-call card showed only "bash" — the bare tool name — instead of what the command does. Fix it by letting each TOOL own how its calls render, rather than the bridge special-casing names. dsh-tools: add an optional two-state presentation seam to ToolDefinition / defineTool — `presentCall(args)` (pending: title, kind, rawInput) and `presentResult(args, result)` (completed: title?, content?). Provider-neutral `ToolCallKind`/`ToolCallPresentation`/`ToolResultPresentation` vocabulary so tools never depend on ACP. defineTool soft-validates args (display runs on log replay, so a malformed/old shape returns undefined instead of throwing). dsh-tool-bash: bash declares presentCall (model `description` → title, exact `command` → rawInput, kind execute) and presentResult (wrap output in a fenced ```console block — a UI-only affordance kept out of the model-facing result); bash_output/bash_kill present task-scoped titles. dsh-acp: inject `tools`; a per-session `ToolPresenter` looks the tool up by name and maps its neutral presentation to the ACP tool_call/tool_call_update wire shape, with a generic fallback (title = name) for tools that declare nothing. Because the `tool/result` event carries only {callId, content, isError}, the presenter keeps a small bridge-local map of ONLY in-flight calls' (name, args), keyed by callId and removed as each result is presented — no event-schema or core change. Replay uses a throwaway presenter so loaded sessions render identically to live ones. Tests: dsh-tools defineTool presenters (typed args, soft-validate), tool-bash bash/bash_output/bash_kill presenters, acp ToolPresenter (tool-owned mapping, unknown-callId fallback, in-flight-only map), and an end-to-end turn through the bridge. The key-gated e2e now asserts a real bash call's title is the model description (not "bash") and rawInput is the command — verified against the real DeepSeek model. The test harness derives its inject from the bridge's exported `inject` so it can't drift again.
2026-06-18 09:01:36 +08:00
/**
* Present foreground calls as terminals and background starts as generic cards.
* The command remains the title on both paths; foreground cwd is passed through
* for the bridge to resolve, while background descriptions remain card content.
feat(acp): tool-owned tool-call UI presentation (title/command/output) In Zed the tool-call card showed only "bash" — the bare tool name — instead of what the command does. Fix it by letting each TOOL own how its calls render, rather than the bridge special-casing names. dsh-tools: add an optional two-state presentation seam to ToolDefinition / defineTool — `presentCall(args)` (pending: title, kind, rawInput) and `presentResult(args, result)` (completed: title?, content?). Provider-neutral `ToolCallKind`/`ToolCallPresentation`/`ToolResultPresentation` vocabulary so tools never depend on ACP. defineTool soft-validates args (display runs on log replay, so a malformed/old shape returns undefined instead of throwing). dsh-tool-bash: bash declares presentCall (model `description` → title, exact `command` → rawInput, kind execute) and presentResult (wrap output in a fenced ```console block — a UI-only affordance kept out of the model-facing result); bash_output/bash_kill present task-scoped titles. dsh-acp: inject `tools`; a per-session `ToolPresenter` looks the tool up by name and maps its neutral presentation to the ACP tool_call/tool_call_update wire shape, with a generic fallback (title = name) for tools that declare nothing. Because the `tool/result` event carries only {callId, content, isError}, the presenter keeps a small bridge-local map of ONLY in-flight calls' (name, args), keyed by callId and removed as each result is presented — no event-schema or core change. Replay uses a throwaway presenter so loaded sessions render identically to live ones. Tests: dsh-tools defineTool presenters (typed args, soft-validate), tool-bash bash/bash_output/bash_kill presenters, acp ToolPresenter (tool-owned mapping, unknown-callId fallback, in-flight-only map), and an end-to-end turn through the bridge. The key-gated e2e now asserts a real bash call's title is the model description (not "bash") and rawInput is the command — verified against the real DeepSeek model. The test harness derives its inject from the bridge's exported `inject` so it can't drift again.
2026-06-18 09:01:36 +08:00
*/
fix(acp): address review of the terminal-card alignment (exit parse, background/error, capability snapshot) Codex + an independent review pass found three real defects in the prior commit: 1. parseExitStatus could misreport a SUCCESSFUL command as a failure: a clean exit 0 appends no marker, so output ending in "[exit code: 5]" (no trailing newline) was read as the marker. Anchor the parse to a LEADING newline — renderResult always inserts one before a real marker, so a body that merely ends in marker-like text no longer matches. A narrow residual (a clean exit 0 whose final line is exactly the marker) is inherent to the replay-only-sees- text design and documented; the complete fix (a structured exit on the event) is the RFC's named escape hatch. 2. A run_in_background start and an isError result were rendered as exited terminal cards with a false exit-0 pill. A background start returns a task-id ack (not a streamed terminal) and is no longer marked terminal; an isError result (spawn failure / abort) carries no exit pill. 3. The terminal capability was re-read live on the result path, so a second initialize between a call and its result could desync them (orphan terminal_output or clobbered card). Snapshot the capability per session at creation (SessionRecord.terminalEnabled) so call and result always agree. Also reword the reference-parity claim: keeping the description as a content block in terminal mode is a DELIBERATE divergence (claude-agent-acp drops it). Tests added for each; with-key e2e still green.
2026-06-18 19:35:15 +08:00
type BashCallArgs = { command: string; description: string; workdir?: string; run_in_background?: boolean }
function presentBashCall(args: BashCallArgs): GenericCallView | TerminalCallView {
if (args.run_in_background === true) {
return {
card: 'generic',
title: args.command,
kind: 'execute',
rawInput: args.command,
content: [{ type: 'text', text: args.description }],
}
}
return {
card: 'terminal',
title: args.command,
description: args.description,
...args.workdir !== undefined ? { cwd: args.workdir } : {},
feat(acp): render bash as a terminal card via the _meta convention When the client advertises clientCapabilities._meta.terminal_output (Zed), a bash tool call now renders as a real TERMINAL card — a cwd header + the command + its output — instead of the plain ```console text block. Keeps agent-side dsh-bash execution; rejects the spec's client-side terminal/create (which would bypass sandbox/env-scrub/ownership/cwd). Matches what claude-agent-acp and codex-acp do; wire contract verified against Zed's source. - dsh-tools: a provider-neutral ToolTerminal shape ({ cwd?, output? }) on ToolCallPresentation/ToolResultPresentation — a tool asks "render me as a terminal"; no ACP types leak in. - dsh-tool-bash: bash presentCall marks terminal (cwd from an explicit absolute workdir, else left for the bridge to fill from the session cwd); presentResult carries the output alongside the ```console fallback. - dsh-acp: initialize reads/remembers the _meta.terminal_output capability; streamSessionEventUpdate maps a terminal presentation to content:[{type:'terminal',terminalId}] + _meta.terminal_info on the call and _meta.terminal_output on the update WHEN capable — else the unchanged text path. terminalId is the callId; cwd defaults to the session header. The pure translator gained a TerminalRendering {enabled,cwd} param (off by default). Tests via the REAL tool-bash + bash-local: capability ON -> terminal content + _meta; OFF -> no _meta (text path). The with-key e2e adds a real-model terminal card case (echo over ACP with the capability on). 773 tests, 100% coverage. The exit-status pill (_meta.terminal_exit), live streaming (_meta.terminal_output_delta), and command classification are RFC follow-ups.
2026-06-18 17:25:09 +08:00
}
feat(acp): tool-owned tool-call UI presentation (title/command/output) In Zed the tool-call card showed only "bash" — the bare tool name — instead of what the command does. Fix it by letting each TOOL own how its calls render, rather than the bridge special-casing names. dsh-tools: add an optional two-state presentation seam to ToolDefinition / defineTool — `presentCall(args)` (pending: title, kind, rawInput) and `presentResult(args, result)` (completed: title?, content?). Provider-neutral `ToolCallKind`/`ToolCallPresentation`/`ToolResultPresentation` vocabulary so tools never depend on ACP. defineTool soft-validates args (display runs on log replay, so a malformed/old shape returns undefined instead of throwing). dsh-tool-bash: bash declares presentCall (model `description` → title, exact `command` → rawInput, kind execute) and presentResult (wrap output in a fenced ```console block — a UI-only affordance kept out of the model-facing result); bash_output/bash_kill present task-scoped titles. dsh-acp: inject `tools`; a per-session `ToolPresenter` looks the tool up by name and maps its neutral presentation to the ACP tool_call/tool_call_update wire shape, with a generic fallback (title = name) for tools that declare nothing. Because the `tool/result` event carries only {callId, content, isError}, the presenter keeps a small bridge-local map of ONLY in-flight calls' (name, args), keyed by callId and removed as each result is presented — no event-schema or core change. Replay uses a throwaway presenter so loaded sessions render identically to live ones. Tests: dsh-tools defineTool presenters (typed args, soft-validate), tool-bash bash/bash_output/bash_kill presenters, acp ToolPresenter (tool-owned mapping, unknown-callId fallback, in-flight-only map), and an end-to-end turn through the bridge. The key-gated e2e now asserts a real bash call's title is the model description (not "bash") and rawInput is the command — verified against the real DeepSeek model. The test harness derives its inject from the bridge's exported `inject` so it can't drift again.
2026-06-18 09:01:36 +08:00
}
/**
* Present completed foreground output as a terminal; background acknowledgements
* and execution errors use generic fenced output without an exit-status pill.
feat(acp): tool-owned tool-call UI presentation (title/command/output) In Zed the tool-call card showed only "bash" — the bare tool name — instead of what the command does. Fix it by letting each TOOL own how its calls render, rather than the bridge special-casing names. dsh-tools: add an optional two-state presentation seam to ToolDefinition / defineTool — `presentCall(args)` (pending: title, kind, rawInput) and `presentResult(args, result)` (completed: title?, content?). Provider-neutral `ToolCallKind`/`ToolCallPresentation`/`ToolResultPresentation` vocabulary so tools never depend on ACP. defineTool soft-validates args (display runs on log replay, so a malformed/old shape returns undefined instead of throwing). dsh-tool-bash: bash declares presentCall (model `description` → title, exact `command` → rawInput, kind execute) and presentResult (wrap output in a fenced ```console block — a UI-only affordance kept out of the model-facing result); bash_output/bash_kill present task-scoped titles. dsh-acp: inject `tools`; a per-session `ToolPresenter` looks the tool up by name and maps its neutral presentation to the ACP tool_call/tool_call_update wire shape, with a generic fallback (title = name) for tools that declare nothing. Because the `tool/result` event carries only {callId, content, isError}, the presenter keeps a small bridge-local map of ONLY in-flight calls' (name, args), keyed by callId and removed as each result is presented — no event-schema or core change. Replay uses a throwaway presenter so loaded sessions render identically to live ones. Tests: dsh-tools defineTool presenters (typed args, soft-validate), tool-bash bash/bash_output/bash_kill presenters, acp ToolPresenter (tool-owned mapping, unknown-callId fallback, in-flight-only map), and an end-to-end turn through the bridge. The key-gated e2e now asserts a real bash call's title is the model description (not "bash") and rawInput is the command — verified against the real DeepSeek model. The test harness derives its inject from the bridge's exported `inject` so it can't drift again.
2026-06-18 09:01:36 +08:00
*/
function presentBashResult(args: unknown, result: ToolResult): ToolResultView | undefined {
feat(acp): tool-owned tool-call UI presentation (title/command/output) In Zed the tool-call card showed only "bash" — the bare tool name — instead of what the command does. Fix it by letting each TOOL own how its calls render, rather than the bridge special-casing names. dsh-tools: add an optional two-state presentation seam to ToolDefinition / defineTool — `presentCall(args)` (pending: title, kind, rawInput) and `presentResult(args, result)` (completed: title?, content?). Provider-neutral `ToolCallKind`/`ToolCallPresentation`/`ToolResultPresentation` vocabulary so tools never depend on ACP. defineTool soft-validates args (display runs on log replay, so a malformed/old shape returns undefined instead of throwing). dsh-tool-bash: bash declares presentCall (model `description` → title, exact `command` → rawInput, kind execute) and presentResult (wrap output in a fenced ```console block — a UI-only affordance kept out of the model-facing result); bash_output/bash_kill present task-scoped titles. dsh-acp: inject `tools`; a per-session `ToolPresenter` looks the tool up by name and maps its neutral presentation to the ACP tool_call/tool_call_update wire shape, with a generic fallback (title = name) for tools that declare nothing. Because the `tool/result` event carries only {callId, content, isError}, the presenter keeps a small bridge-local map of ONLY in-flight calls' (name, args), keyed by callId and removed as each result is presented — no event-schema or core change. Replay uses a throwaway presenter so loaded sessions render identically to live ones. Tests: dsh-tools defineTool presenters (typed args, soft-validate), tool-bash bash/bash_output/bash_kill presenters, acp ToolPresenter (tool-owned mapping, unknown-callId fallback, in-flight-only map), and an end-to-end turn through the bridge. The key-gated e2e now asserts a real bash call's title is the model description (not "bash") and rawInput is the command — verified against the real DeepSeek model. The test harness derives its inject from the bridge's exported `inject` so it can't drift again.
2026-06-18 09:01:36 +08:00
const block = result.content.length === 1 ? result.content[0] : undefined
if (block === undefined || block.type !== 'text') return undefined
const raw = block.text
fix(acp): address review of the terminal-card alignment (exit parse, background/error, capability snapshot) Codex + an independent review pass found three real defects in the prior commit: 1. parseExitStatus could misreport a SUCCESSFUL command as a failure: a clean exit 0 appends no marker, so output ending in "[exit code: 5]" (no trailing newline) was read as the marker. Anchor the parse to a LEADING newline — renderResult always inserts one before a real marker, so a body that merely ends in marker-like text no longer matches. A narrow residual (a clean exit 0 whose final line is exactly the marker) is inherent to the replay-only-sees- text design and documented; the complete fix (a structured exit on the event) is the RFC's named escape hatch. 2. A run_in_background start and an isError result were rendered as exited terminal cards with a false exit-0 pill. A background start returns a task-id ack (not a streamed terminal) and is no longer marked terminal; an isError result (spawn failure / abort) carries no exit pill. 3. The terminal capability was re-read live on the result path, so a second initialize between a call and its result could desync them (orphan terminal_output or clobbered card). Snapshot the capability per session at creation (SessionRecord.terminalEnabled) so call and result always agree. Also reword the reference-parity claim: keeping the description as a content block in terminal mode is a DELIBERATE divergence (claude-agent-acp drops it). Tests added for each; with-key e2e still green.
2026-06-18 19:35:15 +08:00
const isBackground = typeof args === 'object' && args !== null && (args as { run_in_background?: unknown }).run_in_background === true
// Background acknowledgements and errors have no terminal exit status.
if (isBackground || result.isError) {
return { card: 'generic', content: [{ type: 'text', text: `\`\`\`console\n${raw.replace(/\n+$/, '')}\n\`\`\`` }] }
}
return { card: 'terminal', output: raw, ...parseExitStatus(raw) }
feat(acp): tool-owned tool-call UI presentation (title/command/output) In Zed the tool-call card showed only "bash" — the bare tool name — instead of what the command does. Fix it by letting each TOOL own how its calls render, rather than the bridge special-casing names. dsh-tools: add an optional two-state presentation seam to ToolDefinition / defineTool — `presentCall(args)` (pending: title, kind, rawInput) and `presentResult(args, result)` (completed: title?, content?). Provider-neutral `ToolCallKind`/`ToolCallPresentation`/`ToolResultPresentation` vocabulary so tools never depend on ACP. defineTool soft-validates args (display runs on log replay, so a malformed/old shape returns undefined instead of throwing). dsh-tool-bash: bash declares presentCall (model `description` → title, exact `command` → rawInput, kind execute) and presentResult (wrap output in a fenced ```console block — a UI-only affordance kept out of the model-facing result); bash_output/bash_kill present task-scoped titles. dsh-acp: inject `tools`; a per-session `ToolPresenter` looks the tool up by name and maps its neutral presentation to the ACP tool_call/tool_call_update wire shape, with a generic fallback (title = name) for tools that declare nothing. Because the `tool/result` event carries only {callId, content, isError}, the presenter keeps a small bridge-local map of ONLY in-flight calls' (name, args), keyed by callId and removed as each result is presented — no event-schema or core change. Replay uses a throwaway presenter so loaded sessions render identically to live ones. Tests: dsh-tools defineTool presenters (typed args, soft-validate), tool-bash bash/bash_output/bash_kill presenters, acp ToolPresenter (tool-owned mapping, unknown-callId fallback, in-flight-only map), and an end-to-end turn through the bridge. The key-gated e2e now asserts a real bash call's title is the model description (not "bash") and rawInput is the command — verified against the real DeepSeek model. The test harness derives its inject from the bridge's exported `inject` so it can't drift again.
2026-06-18 09:01:36 +08:00
}
feat(acp): honor per-session cwd — run each ACP session in its own workspace Lifts the RFC 010 § Deferred restriction that the server had to launch in the workspace ("cwd must equal the launch directory"). An editor can now open any project folder, and N concurrent sessions over one connection can each target a different directory. - packages/acp: drop the `cwd === process.cwd()` guard in validateWorkspaceParams (keep "must be absolute" — the cwd becomes the session header / bash workdir), and drop the persisted-cwd-vs-launch-dir check in session/load (a resumed session keeps its original header.cwd, so its bash tools run in its workspace). - packages/tool-bash: the missing link — default the bash workdir to the calling agent's session cwd (`exec.agent.session.header.cwd`) via a new resolveWorkdir helper. An explicit model `workdir` still wins; a relative one resolves against the session cwd. This is the only correct spot for multi-session: N sessions share one ctx.bash executor, so the workdir must come per-call from exec.agent, not executor config. Falls back to the executor default when no session cwd is available (preserves non-ACP behavior). - Trust: the cwd originates from the ACP client (the user's editor) at session/new — same trust level as the old launch dir; no new untrusted-input path. `additionalDirectories` (scope widening / sandbox) stays rejected. - Tests: bridge accepts any absolute cwd + records it on the header; session/load honors the persisted cwd; bash defaults to / resolves relative against the session cwd; two sessions with different cwds each run bash in their own dir; non-absolute cwd still rejected. 100% per-file coverage maintained. - Docs: RFC 010 status + § Deferred cwd bullet marked RESOLVED; acp README adds a Per-session cwd section; tool-bash + example READMEs and e2e comments updated.
2026-06-17 10:01:18 +08:00
/**
* Resolve an explicit workdir first, making a relative one session-workspace-relative;
* otherwise use the filesystem identity of the session cwd and leave executor
* defaulting as the fallback. A resolved sandbox-policy root wins so workdir
* and confinement use the exact same per-call identity.
feat(acp): honor per-session cwd — run each ACP session in its own workspace Lifts the RFC 010 § Deferred restriction that the server had to launch in the workspace ("cwd must equal the launch directory"). An editor can now open any project folder, and N concurrent sessions over one connection can each target a different directory. - packages/acp: drop the `cwd === process.cwd()` guard in validateWorkspaceParams (keep "must be absolute" — the cwd becomes the session header / bash workdir), and drop the persisted-cwd-vs-launch-dir check in session/load (a resumed session keeps its original header.cwd, so its bash tools run in its workspace). - packages/tool-bash: the missing link — default the bash workdir to the calling agent's session cwd (`exec.agent.session.header.cwd`) via a new resolveWorkdir helper. An explicit model `workdir` still wins; a relative one resolves against the session cwd. This is the only correct spot for multi-session: N sessions share one ctx.bash executor, so the workdir must come per-call from exec.agent, not executor config. Falls back to the executor default when no session cwd is available (preserves non-ACP behavior). - Trust: the cwd originates from the ACP client (the user's editor) at session/new — same trust level as the old launch dir; no new untrusted-input path. `additionalDirectories` (scope widening / sandbox) stays rejected. - Tests: bridge accepts any absolute cwd + records it on the header; session/load honors the persisted cwd; bash defaults to / resolves relative against the session cwd; two sessions with different cwds each run bash in their own dir; non-absolute cwd still rejected. 100% per-file coverage maintained. - Docs: RFC 010 status + § Deferred cwd bullet marked RESOLVED; acp README adds a Per-session cwd section; tool-bash + example READMEs and e2e comments updated.
2026-06-17 10:01:18 +08:00
*/
function resolveWorkdir(
modelWorkdir: string | undefined,
exec: { agent?: Agent },
policyWorkspaceRoot?: string,
): string | undefined {
const headerCwd = exec.agent?.session.header.cwd
const sessionCwd = policyWorkspaceRoot ?? (headerCwd === undefined ? undefined : canonicalPath(headerCwd))
feat(acp): honor per-session cwd — run each ACP session in its own workspace Lifts the RFC 010 § Deferred restriction that the server had to launch in the workspace ("cwd must equal the launch directory"). An editor can now open any project folder, and N concurrent sessions over one connection can each target a different directory. - packages/acp: drop the `cwd === process.cwd()` guard in validateWorkspaceParams (keep "must be absolute" — the cwd becomes the session header / bash workdir), and drop the persisted-cwd-vs-launch-dir check in session/load (a resumed session keeps its original header.cwd, so its bash tools run in its workspace). - packages/tool-bash: the missing link — default the bash workdir to the calling agent's session cwd (`exec.agent.session.header.cwd`) via a new resolveWorkdir helper. An explicit model `workdir` still wins; a relative one resolves against the session cwd. This is the only correct spot for multi-session: N sessions share one ctx.bash executor, so the workdir must come per-call from exec.agent, not executor config. Falls back to the executor default when no session cwd is available (preserves non-ACP behavior). - Trust: the cwd originates from the ACP client (the user's editor) at session/new — same trust level as the old launch dir; no new untrusted-input path. `additionalDirectories` (scope widening / sandbox) stays rejected. - Tests: bridge accepts any absolute cwd + records it on the header; session/load honors the persisted cwd; bash defaults to / resolves relative against the session cwd; two sessions with different cwds each run bash in their own dir; non-absolute cwd still rejected. 100% per-file coverage maintained. - Docs: RFC 010 status + § Deferred cwd bullet marked RESOLVED; acp README adds a Per-session cwd section; tool-bash + example READMEs and e2e comments updated.
2026-06-17 10:01:18 +08:00
if (modelWorkdir === undefined) return sessionCwd
if (sessionCwd !== undefined && !isAbsolute(modelWorkdir)) {
return resolvePath(sessionCwd, modelWorkdir)
}
return modelWorkdir
}
2026-07-21 03:08:35 +08:00
/** Detach the executor DTO from readonly seam interfaces into plain JSON data. */
function canonicalBashResult(result: BashRunResult) {
const output = (stream: BashRunResult['stdout']) => ({
text: stream.text,
truncated: stream.truncated,
...stream.spillPath !== undefined ? { spillPath: stream.spillPath } : {},
})
return {
exitCode: result.exitCode,
signal: result.signal,
timedOut: result.timedOut,
aborted: result.aborted,
timeoutMs: result.timeoutMs,
stdout: output(result.stdout),
stderr: output(result.stderr),
...result.sandbox !== undefined ? {
sandbox: {
mode: result.sandbox.mode,
denied: result.sandbox.denied,
...result.sandbox.enforcement !== undefined ? { enforcement: result.sandbox.enforcement } : {},
...result.sandbox.runnerFailed !== undefined ? { runnerFailed: result.sandbox.runnerFailed } : {},
},
} : {},
}
}
/** Canonical background-handle properties shared by the bash output union. */
const BACKGROUND_OUTPUT_PROPERTIES = {
kind: { type: 'string', required: true, const: 'background' },
taskId: { type: 'string', required: true },
} as const
export function apply(ctx: Context, config: Config = {}): void {
const bashEnv = new BashEnvRegistry(ctx, config)
bashEnv.register({
name: 'session-persistence',
variables: {
[DSH_SESSION_JSONL_KEY]: {
description: 'Absolute target path of the current session JSONL when the active persistence backend provides one.',
},
},
resolve(execution) {
const agent = execution.agent
if (agent === undefined) return {}
const location = ctx.get('sessionPersistence')?.locate(agent.session.header)
return location?.kind === 'jsonl' ? { [DSH_SESSION_JSONL_KEY]: location.path } : {}
},
})
const backgroundEnabled = config.enableRunInBackground ?? true
const defaultMode = ctx.bash.sandboxMode
const escalationModes: readonly SandboxMode[] = defaultMode === undefined ? [] : ESCALATION_TARGETS
const sandboxPolicy: SandboxPolicyService | undefined = defaultMode === undefined ? undefined : ctx.get('sandboxPolicy')
if (defaultMode !== undefined && sandboxPolicy === undefined) {
throw new Error('tool-bash: the mounted bash executor confines but ctx.sandboxPolicy is missing')
}
/** Resolve the complete standing policy for this call when a confining executor is mounted. */
const resolveSandboxPolicy = (exec: ToolExecution): SandboxExecutionPolicy | undefined =>
sandboxPolicy?.resolve(exec.agent === undefined ? {} : { session: exec.agent.session })
/**
* Resolve a sandbox-escalation request through `ctx.approval` BEFORE
feat(sandbox): cross-family file sandbox — one policy home, sandboxed fs provider, fs escalation parity Extend SandboxMode enforcement from bash to the filesystem tools, the sandbox RFC's deferred cross-family phase. - dsh-sandbox-policy (new, ctx.sandboxPolicy): the single home for the deployment default mode + workspaceRoot and the per-session override event, renamed bash/sandbox-mode -> sandbox/mode and moved here with its fold/setter. Decouples the bash seam from dsh-session. - dsh-fs-sandbox (new): SandboxedFileSystem extends LocalFileSystem and fences write/edit by the per-call mode (read-only denies, workspace-write contains to the workspace + temp roots via the shared writableRoots, danger passes through); reads pass through. Structured FS_SANDBOX_DENIED; in-lock parent re-canonicalization. A policy fence in trusted code, not a kernel boundary. - dsh-sandbox: the shared escalation kit (writableRoots, the strictly-wider ladder, denial/hint markers, approveEscalation) both tool families use; approveEscalation takes a structural approver so dsh-sandbox gains no approval/agent dependency, and both tools stay duplication-free. - tool-fs: write/edit advertise sandbox_permissions/justification under a confining ctx.fs, map FS_SANDBOX_DENIED to the shared [sandbox: ...] marker, and resolve the same one-approved-wider retry. - examples/acp-agent: composes sandbox-policy + fs-sandbox, drops the gating that disabled the fs stack under confined modes. RFC docs/rfc/implemented/feature/2026-07-14-cross-family-fs-sandbox.md; the old sandbox RFC's In-process/deferred/FAQ sections updated to shipped fact.
2026-07-14 20:05:57 +08:00
* anything executes, delegating the shared fail-closed sequence (strict
* widening, channel resolution, outcome mapping) to
* {@link approveEscalation}. This tool contributes only the composition
* guard (the fields are unadvertised without a sandboxing executor, yet
* schema validation checks advertised keys only, so an unadvertised
Merge remote-tracking branch 'origin/master' into cross-family-fs-sandbox # Conflicts: # docs/config-catalog.md # docs/cordis-catalog/services.md # docs/module-graph.md # docs/rfc/implemented/feature/2026-07-06-sandbox.md # examples/acp-agent/tests/snapshots/advanced-toolchain/system-prompt.golden.md # examples/acp-agent/tests/snapshots/advanced-toolchain/tool-schemas.golden.json # examples/acp-agent/tests/snapshots/both-mode-turn/system-prompt.golden.md # examples/acp-agent/tests/snapshots/both-mode-turn/tool-schemas.golden.json # examples/acp-agent/tests/snapshots/code-mode-turn/system-prompt.golden.md # examples/acp-agent/tests/snapshots/escalation-approved/session.jsonl # examples/acp-agent/tests/snapshots/escalation-rejected/session.jsonl # examples/acp-agent/tests/snapshots/hook-cc-pretool-ask/session.jsonl # examples/acp-agent/tests/snapshots/permission-switching/session.jsonl # examples/acp-agent/tests/snapshots/permission-switching/system-prompt.golden.md # examples/acp-agent/tests/snapshots/permission-switching/tool-schemas.golden.json # examples/acp-agent/tests/snapshots/skill-load/tool-schemas.golden.json # examples/acp-agent/tests/snapshots/text-turn/tool-schemas.golden.json # examples/acp-agent/tests/snapshots/workspace-edit/system-prompt.golden.md # examples/acp-agent/tests/snapshots/workspace-edit/tool-schemas.golden.json # packages/bash/bash-sandbox/src/index.ts # packages/bash/bash-sandbox/tests/bwrap.e2e.ts # packages/bash/bash-sandbox/tests/sandbox.spec.ts # packages/bash/bash-sandbox/tests/seatbelt.e2e.ts # packages/bash/bash/src/index.ts # packages/bash/tool-bash/package.json # packages/bash/tool-bash/src/index.ts # packages/bash/tool-bash/src/render.ts # packages/bash/tool-bash/tests/tools.spec.ts # packages/bash/tool-bash/tsconfig.json # pnpm-lock.yaml # scripts/verify-package-readme-model-experience.ts
2026-07-16 23:31:51 +08:00
* `sandbox_permissions` still reaches execute) and the approval ingredients
* The shared policy resolver is required whenever the executor advertises
* confinement, so a split composition fails at tool-plugin load.
*/
const approveBashEscalation = (
mode: string,
justification: string,
exec: ToolExecution,
standingPolicy: SandboxExecutionPolicy | undefined,
): Promise<SandboxMode> => {
if (escalationModes.length === 0) {
throw new Error('sandbox_permissions is not available in this composition (no sandboxing executor to escalate)')
}
const effectiveMode = (standingPolicy as SandboxExecutionPolicy).mode
feat(sandbox): cross-family file sandbox — one policy home, sandboxed fs provider, fs escalation parity Extend SandboxMode enforcement from bash to the filesystem tools, the sandbox RFC's deferred cross-family phase. - dsh-sandbox-policy (new, ctx.sandboxPolicy): the single home for the deployment default mode + workspaceRoot and the per-session override event, renamed bash/sandbox-mode -> sandbox/mode and moved here with its fold/setter. Decouples the bash seam from dsh-session. - dsh-fs-sandbox (new): SandboxedFileSystem extends LocalFileSystem and fences write/edit by the per-call mode (read-only denies, workspace-write contains to the workspace + temp roots via the shared writableRoots, danger passes through); reads pass through. Structured FS_SANDBOX_DENIED; in-lock parent re-canonicalization. A policy fence in trusted code, not a kernel boundary. - dsh-sandbox: the shared escalation kit (writableRoots, the strictly-wider ladder, denial/hint markers, approveEscalation) both tool families use; approveEscalation takes a structural approver so dsh-sandbox gains no approval/agent dependency, and both tools stay duplication-free. - tool-fs: write/edit advertise sandbox_permissions/justification under a confining ctx.fs, map FS_SANDBOX_DENIED to the shared [sandbox: ...] marker, and resolve the same one-approved-wider retry. - examples/acp-agent: composes sandbox-policy + fs-sandbox, drops the gating that disabled the fs stack under confined modes. RFC docs/rfc/implemented/feature/2026-07-14-cross-family-fs-sandbox.md; the old sandbox RFC's In-process/deferred/FAQ sections updated to shipped fact.
2026-07-14 20:05:57 +08:00
return approveEscalation(
{ requestedMode: mode, justification, effectiveMode, subject: 'command' },
{
approver: ctx.get('approval'),
agent: exec.agent,
callId: exec.callId,
toolName: 'bash',
signal: exec.signal,
feat(sandbox): cross-family file sandbox — one policy home, sandboxed fs provider, fs escalation parity Extend SandboxMode enforcement from bash to the filesystem tools, the sandbox RFC's deferred cross-family phase. - dsh-sandbox-policy (new, ctx.sandboxPolicy): the single home for the deployment default mode + workspaceRoot and the per-session override event, renamed bash/sandbox-mode -> sandbox/mode and moved here with its fold/setter. Decouples the bash seam from dsh-session. - dsh-fs-sandbox (new): SandboxedFileSystem extends LocalFileSystem and fences write/edit by the per-call mode (read-only denies, workspace-write contains to the workspace + temp roots via the shared writableRoots, danger passes through); reads pass through. Structured FS_SANDBOX_DENIED; in-lock parent re-canonicalization. A policy fence in trusted code, not a kernel boundary. - dsh-sandbox: the shared escalation kit (writableRoots, the strictly-wider ladder, denial/hint markers, approveEscalation) both tool families use; approveEscalation takes a structural approver so dsh-sandbox gains no approval/agent dependency, and both tools stay duplication-free. - tool-fs: write/edit advertise sandbox_permissions/justification under a confining ctx.fs, map FS_SANDBOX_DENIED to the shared [sandbox: ...] marker, and resolve the same one-approved-wider retry. - examples/acp-agent: composes sandbox-policy + fs-sandbox, drops the gating that disabled the fs stack under confined modes. RFC docs/rfc/implemented/feature/2026-07-14-cross-family-fs-sandbox.md; the old sandbox RFC's In-process/deferred/FAQ sections updated to shipped fact.
2026-07-14 20:05:57 +08:00
},
)
}
// Cross-call guidance belongs in the prompt rather than one-call schema prose.
ctx.systemPrompt.section({
name: 'tool:bash',
order: 105,
text: 'Check the [exit code: N] marker on every bash result; investigate failures before moving on.',
})
ctx.tools.register(defineTool({
name: 'bash',
description: bashDescription(backgroundEnabled, escalationModes),
parameters: {
command: { type: 'string', required: true, description: 'The bash command to execute.' },
description: {
type: 'string',
required: true,
description: 'Clear, concise description of what this command does in active voice, '
+ '5-10 words (shown in the UI). Examples: "ls" → "List files in current directory"; '
+ '"git status" → "Show working tree status"; "npm install" → "Install package dependencies".',
},
timeoutMs: { type: 'number', description: 'Timeout in milliseconds. The executor applies its configured default and cap, and kills the command on expiry.' },
feat(acp): honor per-session cwd — run each ACP session in its own workspace Lifts the RFC 010 § Deferred restriction that the server had to launch in the workspace ("cwd must equal the launch directory"). An editor can now open any project folder, and N concurrent sessions over one connection can each target a different directory. - packages/acp: drop the `cwd === process.cwd()` guard in validateWorkspaceParams (keep "must be absolute" — the cwd becomes the session header / bash workdir), and drop the persisted-cwd-vs-launch-dir check in session/load (a resumed session keeps its original header.cwd, so its bash tools run in its workspace). - packages/tool-bash: the missing link — default the bash workdir to the calling agent's session cwd (`exec.agent.session.header.cwd`) via a new resolveWorkdir helper. An explicit model `workdir` still wins; a relative one resolves against the session cwd. This is the only correct spot for multi-session: N sessions share one ctx.bash executor, so the workdir must come per-call from exec.agent, not executor config. Falls back to the executor default when no session cwd is available (preserves non-ACP behavior). - Trust: the cwd originates from the ACP client (the user's editor) at session/new — same trust level as the old launch dir; no new untrusted-input path. `additionalDirectories` (scope widening / sandbox) stays rejected. - Tests: bridge accepts any absolute cwd + records it on the header; session/load honors the persisted cwd; bash defaults to / resolves relative against the session cwd; two sessions with different cwds each run bash in their own dir; non-absolute cwd still rejected. 100% per-file coverage maintained. - Docs: RFC 010 status + § Deferred cwd bullet marked RESOLVED; acp README adds a Per-session cwd section; tool-bash + example READMEs and e2e comments updated.
2026-06-17 10:01:18 +08:00
workdir: { type: 'string', description: 'Working directory for this command. Defaults to the session workspace; a relative path is resolved against it.' },
...backgroundEnabled ? {
run_in_background: { type: 'boolean' as const, description: 'Run in the background and return a task id immediately (collect with task_output, stop with task_kill). No timeout applies.' },
} : {},
...escalationModes.length > 0 ? {
sandbox_permissions: {
type: 'string' as const,
enum: [...escalationModes],
description: 'The wider sandbox mode this command needs. Only valid as a one-shot retry of a command the sandbox just denied; requires justification and user approval.',
},
justification: {
type: 'string' as const,
description: 'Required with sandbox_permissions: one sentence for the user explaining why this exact command needs the wider access.',
},
} : {},
},
2026-07-21 03:08:35 +08:00
output: {
schema: {
oneOf: [
{
type: 'object',
additionalProperties: false,
properties: BACKGROUND_OUTPUT_PROPERTIES,
},
{
type: 'object',
additionalProperties: false,
properties: {
kind: { type: 'string', required: true, const: 'foreground' },
exitCode: { required: true, oneOf: [{ type: 'integer' }, { type: 'null' }] },
signal: { required: true, oneOf: [{ type: 'string' }, { type: 'null' }] },
timedOut: { type: 'boolean', required: true },
aborted: { type: 'boolean', required: true },
timeoutMs: { type: 'number', required: true },
stdout: {
type: 'object',
additionalProperties: false,
required: true,
properties: {
text: { type: 'string', required: true },
truncated: { type: 'boolean', required: true },
spillPath: { type: 'string' },
},
},
stderr: {
type: 'object',
additionalProperties: false,
required: true,
properties: {
text: { type: 'string', required: true },
truncated: { type: 'boolean', required: true },
spillPath: { type: 'string' },
},
},
sandbox: {
type: 'object',
additionalProperties: false,
properties: {
mode: { type: 'string', required: true },
denied: { type: 'boolean', required: true },
enforcement: { type: 'string' },
runnerFailed: { type: 'boolean' },
},
},
},
},
],
},
render: (_args, value) => [{
type: 'text',
text: value.kind === 'background'
? `started background task ${value.taskId}`
: renderResult(value as { kind: 'foreground' } & BashRunResult, escalationModes),
}],
},
async execute(args: BashToolArgs, exec) {
validateBashArgs(args)
// Description is display metadata; workdir defaults to the caller's session.
const standingPolicy = resolveSandboxPolicy(exec)
const approvedMode = args.sandbox_permissions !== undefined && args.justification !== undefined
? await approveBashEscalation(args.sandbox_permissions, args.justification, exec, standingPolicy)
: undefined
const policy = approvedMode === undefined
? standingPolicy
: { ...(standingPolicy as SandboxExecutionPolicy), mode: approvedMode }
const workdir = resolveWorkdir(args.workdir, exec, standingPolicy?.workspaceRoot)
const dshEnv = bashEnv.collect(exec)
const request = {
command: args.command,
feat(acp): honor per-session cwd — run each ACP session in its own workspace Lifts the RFC 010 § Deferred restriction that the server had to launch in the workspace ("cwd must equal the launch directory"). An editor can now open any project folder, and N concurrent sessions over one connection can each target a different directory. - packages/acp: drop the `cwd === process.cwd()` guard in validateWorkspaceParams (keep "must be absolute" — the cwd becomes the session header / bash workdir), and drop the persisted-cwd-vs-launch-dir check in session/load (a resumed session keeps its original header.cwd, so its bash tools run in its workspace). - packages/tool-bash: the missing link — default the bash workdir to the calling agent's session cwd (`exec.agent.session.header.cwd`) via a new resolveWorkdir helper. An explicit model `workdir` still wins; a relative one resolves against the session cwd. This is the only correct spot for multi-session: N sessions share one ctx.bash executor, so the workdir must come per-call from exec.agent, not executor config. Falls back to the executor default when no session cwd is available (preserves non-ACP behavior). - Trust: the cwd originates from the ACP client (the user's editor) at session/new — same trust level as the old launch dir; no new untrusted-input path. `additionalDirectories` (scope widening / sandbox) stays rejected. - Tests: bridge accepts any absolute cwd + records it on the header; session/load honors the persisted cwd; bash defaults to / resolves relative against the session cwd; two sessions with different cwds each run bash in their own dir; non-absolute cwd still rejected. 100% per-file coverage maintained. - Docs: RFC 010 status + § Deferred cwd bullet marked RESOLVED; acp README adds a Per-session cwd section; tool-bash + example READMEs and e2e comments updated.
2026-06-17 10:01:18 +08:00
...workdir !== undefined ? { workdir } : {},
...args.timeoutMs !== undefined ? { timeoutMs: args.timeoutMs } : {},
dshEnv,
...policy !== undefined ? { sandboxPolicy: policy } : {},
}
if (args.run_in_background === true) {
// Undeclared keys are allowed, so schema omission also needs enforcement.
if (!backgroundEnabled) {
throw new Error('run_in_background is disabled for this deployment (enableRunInBackground: false)')
}
const tasks = ctx.get('tasks')
if (tasks === undefined) {
throw new Error('background tasks unavailable: load @deepseek-ai/dsh-tasks and @deepseek-ai/dsh-tool-tasks')
}
// The caller owns cancellation until ctx.tasks commits detached ownership.
if (exec.signal.aborted) {
const error = new HarnessError('tool call aborted', TOOL_ABORTED)
error.name = 'AbortError'
throw error
}
// Task preflight finishes before the starter can spawn a process.
const id = tasks.start({
kind: 'bash',
label: args.command,
...exec.agent ? { owner: exec.agent } : {},
run: () => {
const proc = ctx.bash.start(ctx.bash.resolve(request))
return {
cancel: () => void proc.kill(),
done: proc.done.then(() => processOutcome(proc)),
readOutput: () => renderProcessRead(proc.readOutput(), proc.sandbox, escalationModes),
}
},
})
2026-07-21 03:08:35 +08:00
return { kind: 'background' as const, taskId: id }
}
const result = await ctx.bash.run(ctx.bash.resolve({
...request,
signal: exec.signal,
}))
if (result.aborted) throw new Error('command aborted')
2026-07-21 03:08:35 +08:00
return { kind: 'foreground' as const, ...canonicalBashResult(result) }
},
feat(acp): tool-owned tool-call UI presentation (title/command/output) In Zed the tool-call card showed only "bash" — the bare tool name — instead of what the command does. Fix it by letting each TOOL own how its calls render, rather than the bridge special-casing names. dsh-tools: add an optional two-state presentation seam to ToolDefinition / defineTool — `presentCall(args)` (pending: title, kind, rawInput) and `presentResult(args, result)` (completed: title?, content?). Provider-neutral `ToolCallKind`/`ToolCallPresentation`/`ToolResultPresentation` vocabulary so tools never depend on ACP. defineTool soft-validates args (display runs on log replay, so a malformed/old shape returns undefined instead of throwing). dsh-tool-bash: bash declares presentCall (model `description` → title, exact `command` → rawInput, kind execute) and presentResult (wrap output in a fenced ```console block — a UI-only affordance kept out of the model-facing result); bash_output/bash_kill present task-scoped titles. dsh-acp: inject `tools`; a per-session `ToolPresenter` looks the tool up by name and maps its neutral presentation to the ACP tool_call/tool_call_update wire shape, with a generic fallback (title = name) for tools that declare nothing. Because the `tool/result` event carries only {callId, content, isError}, the presenter keeps a small bridge-local map of ONLY in-flight calls' (name, args), keyed by callId and removed as each result is presented — no event-schema or core change. Replay uses a throwaway presenter so loaded sessions render identically to live ones. Tests: dsh-tools defineTool presenters (typed args, soft-validate), tool-bash bash/bash_output/bash_kill presenters, acp ToolPresenter (tool-owned mapping, unknown-callId fallback, in-flight-only map), and an end-to-end turn through the bridge. The key-gated e2e now asserts a real bash call's title is the model description (not "bash") and rawInput is the command — verified against the real DeepSeek model. The test harness derives its inject from the bridge's exported `inject` so it can't drift again.
2026-06-18 09:01:36 +08:00
presentCall: presentBashCall,
presentResult: presentBashResult,
}))
}