The web task list gains a two-press stop control on running job rows, built on
one seam change: JobRegistry.kill now takes { reason?, reported? } instead of a
positional reason. reported defaults to true (the model's job_kill claims the
terminal report exactly as before); the web path passes reported: false because
a human kill has no model-visible channel, so the settlement flows through the
existing tool-jobs completion notice (wakeup/inject, budgets, truncation all
unchanged). A recorded kill reason merges into a killed settlement's detail
(producer facts first: 'signal: SIGTERM; cancelled by the user'), so the notice
and the web row both say who stopped the work.
- session-controller: session.killJob Remote beside cancel, live-only Agent
lookup, job-not-found/jobs-unavailable rejections joined to the closed RPC
error table; ClientSession.killJob and the ISession face carry it to the UI.
- ui-activity: arm/confirm stop button (3s disarm, pending disable, failure
hint), rendered only on running job rows; standalone activities keep no kill
handle and the README limitation now says so.
- The background-job-list web e2e now drives the whole human path through the
button; goldens show the merged killed detail end to end.
- Agent Note 2026-08-26-human-job-kill records the decision against Kimi's
suppress-own-notification precedent and Claude Code's silent-kill failure;
the 2026-08-08 blocker note and the unified-task-list note point at it.
description, kind
| description | kind |
|---|---|
| The bash capability family for deployments and maintainers choosing and composing a shell executor, sandboxing, and the model-facing bash and pwsh tools. | package-group |
shell/ — bash capability family
English | 中文
Summary
The shell group provides command execution to agents: run a foreground command and read its bounded output, or start a background process and poll it, on POSIX with Bash and on Windows with PowerShell. Exactly one executor implementation is mounted per composition; the sandboxing executors confine every command through the sandbox capability, and the model-facing bash and pwsh tools sit on top of whichever executor is mounted. Choose a Bash executor for POSIX, a PowerShell executor for Windows, and pick the sandboxing variant when commands need file-level confinement.
Table of Contents
Packages
| Package | Role | ctx key |
|---|---|---|
shell |
Defines the executor contract: foreground runs, background handles, and request resolution | ctx.shell |
bash-local |
Runs Bash commands as fresh bash -c processes on POSIX |
registers ctx.shell |
bash-sandbox |
Runs Bash commands confined through the sandbox capability, reporting denials as facts | registers ctx.shell |
pwsh-local |
Runs PowerShell commands as fresh pwsh -Command processes on Windows |
registers ctx.shell |
pwsh-sandbox |
Runs PowerShell commands confined through the sandbox capability | registers ctx.shell |
shell-env |
Supplies the managed DSH_* environment every shell command receives |
ctx.shellEnv |
tool-bash |
Exposes Bash execution and background jobs to the model as the bash tool |
registers on ctx.tools |
tool-bash-persistent |
Runs model shell calls in one owner-isolated persistent Bash session | registers on ctx.tools |
tool-pwsh |
Exposes PowerShell execution to the model as the pwsh tool |
registers on ctx.tools |
tool-pwsh-persistent |
Runs model shell calls in one owner-isolated persistent PowerShell session | registers on ctx.tools |
A profile layer selects exactly one executor implementation (the win32 layer swaps the POSIX rows for the pwsh ones; mounting two fails loud on the duplicate service registration) and the model-facing tools it needs. A sandboxed composition also selects a ctx.sandbox provider and ctx.sandboxPolicy; the base bundle owns the shipped wiring.
Related documentation
- Bash executor subsystem — the shared request/spec vocabulary, results, background processes, and the service contract.
- Sandbox subsystem — the confinement capability the sandboxing executors consume.
Dev Note
None.