Files
deepseek-harness/packages/subagent/subagent-acp
Tianyi Cui f256f3961d feat(system-prompt): prompt variables, persona-as-section, tool-guidance ownership
One principle: every fact in the assembled prompt has exactly one owner.

- dsh-system-prompt: merge-extensible AssembleContext on assemble();
  a variable(name, provider) registry; {{name}} interpolation in
  renderPrompt, strict (unknown/valueless/malformed references throw);
  duplicate section and variable names rejected; assembly carries
  resolved section text + variables through the assemble waterfall.
- dsh-agent declares AssembleContext.agent; dsh-agent-loop registers
  the agent:persona section (order 0 - identity renders before tool
  guidance) and the model/cwd variables, and drops its string join:
  renderPrompt(assembly) IS the full prompt.
- Tool guidance moves to its owners: descriptions carry per-tool
  semantics; sections only cross-call habits (tool:bash exit-code
  habit at order 105; read's not-shell nudge). todo/subagent need no
  section - their descriptions already carry the contract.
- SubagentProvider.inheritsParentContext (spawn/acp false, fork true);
  dsh-tool-subagent derives truthful per-provider wording and resolves
  the provider at load (backend must be listed first).
- Example personas shrink to identity + behavior with {{model}} (and
  {{cwd}} in the ACP tree); the welcome banner stops enumerating tools.

RFC: docs/rfc/implemented/architecture/2026-07-05-prompt-variables-and-tool-guidance-ownership.md
2026-07-05 01:54:46 +08:00
..

@deepseek-ai/dsh-subagent-acp

The out-of-process ACP subagent backend: runs each child agent in a spawned subprocess, driven over the Agent Client Protocol (ACP) as the client. Registers a SubagentProvider on ctx.subagents (the subagent seam), alongside the in-process -spawn/-fork backends — multiple backends coexist by name.

It is the direction-inverted twin of the server-side bridge in @deepseek-ai/dsh-acp: that package is the ACP agent (it answers initialize/newSession/prompt); this one is the ACP client (it calls them and implements the sessionUpdate/requestPermission callbacks). Point the configured command at the acp-agent example to "talk to our own process".

What it does

start(request) spawns the configured command, wraps its stdio in an ACP ClientSideConnection, and drives one session: initializenewSessionprompt. The child's streamed agent_message_chunk text becomes the SubagentResult.output; the prompt's terminal StopReason maps to the stop reason. dispose() kills the subprocess and awaits its exit.

Fresh process per run. Each start spawns a new child, runs exactly one ACP session, and disposes it. Persistent-process pooling is a future optimization (see the RFC).

Unlike the in-process backends, the child does NOT share this cordis context — it is a separate process with its own session, model client, and tools. So this backend:

  • injects only subagents (no ctx.agents);
  • advertises NO start-time capabilities (an out-of-process child can't enforce the parent's depth/tool-filter);
  • ignores request.parent.

Config

Key Type Default Notes
providerName string acp Registry name on ctx.subagents.
command string — (required) The executable to spawn for each run (the child ACP agent).
args string[] [] Arguments passed to command.
cwd string parent cwd Working directory for the child process and its ACP session.
permission 'allow' | 'reject' reject How to auto-answer the child's session/request_permission prompts. reject declines every prompt (answer cancelled); allow approves via the first allow-shaped option. The first cut surfaces no prompt to a human.
env Record<string,string> {} Extra env vars for the child (e.g. its own DEEPSEEK_API_KEY). Forwarded on top of a credential-scrubbed copy of the parent env, so an explicit key reaches the child while ambient secrets do not leak implicitly.
disposeEofGraceMs number 6000 Dispose ladder tier 1: how long the child gets to quiesce on its own after stdin EOF (flush persistence, tear down its nested subprocesses) before SIGTERM.
disposeGraceMs number 3000 Dispose ladder tier 2: grace between SIGTERM and the SIGKILL escalation.
- id: subagent-acp
  name: '@deepseek-ai/dsh-subagent-acp'
  config:
    providerName: acp
    command: node
    args: ['--import', 'tsx', './packages/ui/acp-agent/src/bin.ts', './examples/acp-agent/cordis.yml']
    permission: reject
    env:
      DEEPSEEK_API_KEY: !!js process.env.DEEPSEEK_API_KEY

StopReason mapping

ACP StopReason → harness SubagentStopReason:

ACP harness
end_turn completed
max_tokens max-tokens
refusal refusal
cancelled aborted
max_turn_requests error (no clean equivalent; the task did not finish)
(unknown) error

A spawn/transport/RPC failure resolves error (or aborted if a cancel was requested) — result never rejects on a child-level failure, per the seam contract.

Environment scrub

Credential-shaped ambient vars (/KEY|SECRET|TOKEN/i) are NOT forwarded to the child by default — the parent harness's own secrets must not leak into a spawned process implicitly. The child's OWN credentials are supplied explicitly via config.env, layered AFTER the scrub, so an intended DEEPSEEK_API_KEY survives while an incidental AWS_SECRET_ACCESS_KEY does not.

Testing

  • Keyless (subagent-acp.spec.ts): spawns a scripted mock ACP server subprocess (tests/mock-acp-server.ts) and drives it through the real backend over real ACP stdio — connection setup, client callbacks, the prompt round-trip, stop-reason mapping, cancellation (including the early-cancel race and a torn-pipe-after-cancel), permission auto-answer, and quiescent disposal. No model, no key.
  • With-key e2e (subagent-acp.e2e.ts): the harness drives ITSELF — the backend spawns the real acp-agent example process and a real model in that child answers a prompt and does real file work (verified on disk). Self-skips without DEEPSEEK_API_KEY.

TODO(acp-subagent-replay): snapshot-tier coverage of an ACP child is a separate replay shape (each child is its own PROCESS with its own single-agent replay, distinct from the in-process per-session keying), deferred — see the RFC.

Plugin export shape

Named name / inject / Config / apply, with no default export: the cordis Loader's unwrapExports does exports.default ?? exports, so a stray default would collapse the module to the bare function and drop the inject namespace (see docs/postmortem/0001).