f256f3961d
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
@deepseek-ai/dsh-acp-agent
The ACP server app: a Cordis app plugin that composes the providerless agent spine (@deepseek-ai/dsh-agent-core) with the front-door cluster an Agent Client Protocol server needs, and a bin that boots a leaf cordis.yml speaking ACP JSON-RPC on stdio.
It is the structured counterpart to @deepseek-ai/dsh-stdio-agent: both consume the same spine, but this one bakes in the OPPOSITE front-door cluster.
What it bakes in — and what it deliberately omits
stdout is the ACP JSON-RPC channel, so the cluster is defined as much by what it LEAVES OUT as what it includes:
| Plugin | Why |
|---|---|
@deepseek-ai/dsh-agent-core |
the spine, pre-creating no agents (ACP session/new creates them on demand) |
@deepseek-ai/dsh-session-persistence-jsonl |
durable JSONL session log (the bridge advertises loadSession) |
@deepseek-ai/dsh-acp |
the bridge that owns stdout for JSON-RPC |
| omitted — it writes to stdout and would corrupt the protocol frames (the stdout-purity footgun) | |
hmr |
omitted — the editor owns the subprocess |
Because the package wires no logger entry, an ACP leaf has nothing to get wrong by default: it only picks backends, so the common mistake — copying a console-logger entry from the stdio config — has no place here. (A leaf author technically can still add @cordisjs/plugin-logger-console as a sibling entry; the package can't forbid that. So the rule stands: never add a stdout logger to an ACP leaf — stdout is the JSON-RPC channel. Use a stderr exporter if you need logs.)
Config
| Key | Default | Routed to |
|---|---|---|
model |
(required) | the per-session agent template the bridge creates agents from |
systemPrompt |
(required) | the per-session agent's persona template (may reference {{model}}/{{cwd}}) |
persistenceRoot |
./.sessions |
the JSONL backend's root directory |
The leaf supplies the swappable backends: an LLM adapter (llm-deepseek for the real model, llm-replay for keyless snapshot replay) and a bash executor (bash-local).
The bin
dsh-acp-agent [path-to-cordis.yml] (default ./cordis.yml):
- loads a gitignored
.envfrom the cwd — skipped in snapshot REPLAY so a stray key can never trigger a live call; - honors
DSH_SNAPSHOT=replayby booting the siblingcordis.snapshot.yml(the keyless replay tree,llm-replayin place ofllm-deepseek); - in a snapshot run, disposes the context on stdin EOF so the session log is fully flushed before exit.
Run it under node --expose-internals: the cordis Loader resolves the config's bare plugin specifiers through its internal module loader, active only under that flag. (demo:acp runs under tsx, whose tsconfig paths map resolves them instead.)
All diagnostics go to stderr — stdout is the protocol.