Yichen Jiang
1cfbafab6a
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 Three limits are recorded rather than fixed: projection key presence is not a per-session capability signal, a superseded standing generation is never reclaimed, and a `cordis_mount` temporary plugin belongs to the composition rather than the session that mounted it.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
@deepseek-ai/dsh-acp
English | 中文
Automation-only Agent Client Protocol server over JSON-RPC stdio. Programmatic clients create fresh harness agents, send text prompts, collect committed assistant text, resolve one-shot permission requests by policy, and cancel work. The primary in-repository client is dsh-subagent-acp.
This package is a transport adapter, not a UI integration or a capability seam. It does not expose editor navigation, transcript replay, commands, modes, configuration pickers, elicitation, reasoning, plans, titles, or tool presentation. Interactive rendering and human questions belong to the Web host and client modules.
Plugin
apply(ctx, config) opens an AgentSideConnection on stdin/stdout and drives ctx.agents. Stdout is reserved for protocol frames.
| Config | Default | Meaning |
|---|---|---|
provider |
— | Initial provider route for every created agent. |
model |
— | Initial model for every created agent. |
Both fields are optional so another agent/request listener may supply the target. The runnable ACP composition requires both.
Protocol contract
| Method | Behavior |
|---|---|
initialize |
Negotiates the supported version and advertises baseline-only prompts (no image, audio, or embedded-context capability). No session, editor, terminal, filesystem, or MCP capability is advertised. |
authenticate |
No-op because the server advertises no authentication methods. |
session/new |
Creates a fresh agent with an absolute primary cwd; empty additionalDirectories and mcpServers are accepted, non-empty values reject. |
session/prompt |
Concatenates text blocks, renders baseline resource links as bracketed textual references, rejects empty or beyond-baseline input, permits one in-flight request per session, and waits for the whole agent to become idle. Normal quiescence reports end_turn; explicit ACP cancellation, disposal, or a prompt whose admission was discarded (a turnless slot) reports cancelled. |
session/cancel |
Cancels only the addressed agent and settles its pending prompt as cancelled; unknown ids are no-ops. |
session/update |
Emits one agent_message_chunk per non-empty text block in a committed assistant/message. Raw deltas and non-message events are omitted. |
session/request_permission |
Offers one-shot allow/reject choices for bridge-owned approval requests carrying a tool call id. Clients may answer automatically. |
One connection may own several sessions. The bridge keys records by branded session id and checks exact agent identity before routing events or permission requests. Each session has an independent prompt slot, workspace, cancellation path, and disposer.
Committed-message output intentionally trades token-by-token latency for a clean automation result. Uncommitted provider chunks and retry attempts cannot leak partial text; reasoning and tool activity remain in the session log for observability through other interfaces.
Lifecycle
Client disconnect and Cordis disposal share one memoized teardown. The bridge first rejects new sessions and prompts, settles pending prompts, then drains continuable descendants only below this connection's exact owned Agents before disposing those handles in parallel and awaiting every result before reporting any failure. Other frontends sharing the Context retain their continuable forests and admission. An ACP-only plugin reload therefore leaves no orphan agent.
ACP requires each prompt response to carry a stopReason, but the bridge does not claim a prompt-specific turn outcome. Committed assistant messages stream across the owned activity, and steering or injected work may contribute before idle. Token-limit turn endings therefore do not become prompt-level ACP stop reasons (they settle as end_turn); a model error on the correlated turn rejects the prompt immediately.
Running
pnpm --dir /path/to/deepseek-harness run demo:acp boots the repository's automation server composition. A parent harness can spawn it through @deepseek-ai/dsh-subagent-acp; other ACP clients need only the core methods above.
Model Experience
Prompt text
What the model sees
session/prompt text blocks are concatenated verbatim into one user message; a baseline resource link appears in that message as a bracketed [resource_link name=… uri=…] reference the model may open with its own tools. Protocol metadata, client capabilities, permission choices, and session ids never enter the model request.
Token effect
Prompt tokens are data-dependent and remain in that session's history until compaction. Concurrent ACP sessions retain independent contexts.
KV Cache effect
Append-only; the new user message follows the reusable request prefix and does not invalidate prior cache entries.
Permission decisions
What the model sees
Nothing directly. The owning tool records its allowed, rejected, cancelled, or unavailable outcome through the normal tool-result path.
Token effect
Only the owning tool result contributes tokens.
KV Cache effect
Append-only through the owning tool result.
Known Limitations and Deferred Work
- Fresh sessions only — load, list, resume, delete, and fork are unsupported.
- Baseline prompts and one workspace only — images, audio, embedded resources, non-empty additional directories, and MCP servers reject; resource links flatten to textual references rather than fetched content.
- Committed answers only — live progress, reasoning, tool activity, plans, titles, and usage stay off the wire.
- Connection-owned lifetime — one connection releases all of its sessions; per-session close is not implemented.