deepseek-harness/packages/core/agent-loop/README.md

description, kind

description	kind
The default agent driver for users and maintainers choosing, configuring, or debugging how agents are created and how turns and steps run.	package-reference

@deepseek-ai/dsh-agent-loop

English | 中文

Summary

dsh-agent-loop creates fresh agents or resumes persisted sessions, then drives each turn through model requests, streamed responses, tool execution, and durable session history. Mount it for standard agent compositions; declarative entries start agents at boot, while the public ctx.agents API supports programmatic creation and resume. maxParallelToolCalls limits concurrent parallel-safe calls, and exclusive calls retain ordering. Cancellation preserves streamed text already delivered to the user. Choose a custom Agent implementation only when the standard "call model, run tools, repeat" lifecycle is insufficient.

Table of Contents

• Use this package
• Understand the implementation
• Further Exploration
• Model Experience
• Known Limitations and Deferred Work
• Dev Note

---

Use this package

Mount dsh-agent-loop in any composition that should run agents. It supplies the driver behind ctx.agents and starts any agents you declare in its config; both dsh-base and dsh-sdk-minimal mount it as an explicit row.

Configure declarative agents

Agents declared in the config start automatically when the plugin loads. Each entry needs an id label; a model call additionally requires both provider and model (agent/request may supply a missing pair before dispatch).

- name: '@deepseek-ai/dsh-agent-loop'
  config:
    maxParallelToolCalls: 10
    agents:
      - id: 'main'
        provider: deepseek
        model: deepseek-chat
        reasoningEffort: high
        cwd: /workspace

Field	Default	Meaning
maxParallelToolCalls	10	Parallel-safe tool calls in flight per step; 1 is serial
agents[].id	required	Stable label; a fresh session mints ${id}-session-<uuid> unless sessionId is set
agents[].provider / agents[].model	—	Model route; both required before dispatch
agents[].reasoningEffort	—	Non-empty initial reasoning effort; agent/request may override it
agents[].maxTokens	—	Positive per-request output-token cap
agents[].cwd	—	Workspace directory for a fresh session
agents[].sessionId	—	Exact identity: first use creates, a remount resumes materialized history
agents[].resumeSessionId	—	Load this persisted session instead of creating one; mutually exclusive with sessionId

The generated configuration catalog is the exhaustive source for every accepted field. The adapter validates the effective reasoning effort and the loop records it in the request header. maxParallelToolCalls is also the whole agent-loop settings section, so a user layer over this entry caps the next tool group without a restart.

Create or resume agents programmatically

Plugins and hosts create agents through ctx.agents.create() and resume persisted sessions through ctx.agents.resume(); both return an AgentHandle whose dispose() owns exact teardown. The loop runs each created agent to completion — the handle is only needed when the caller must tear the agent down itself.

const handle = await ctx.agents.create({
  sessionId,
  agentOptions: { provider: 'deepseek', model: 'deepseek-chat' },
  setup: (agentCtx) => { /* scoped tools, prompt sections, listeners */ },
})

Every inbox mutation commits one normalized agent/inbox/spliced event. The projection registry folds that event synchronously, so the live projection reflects the splice when Session.append() returns. Insertions, edits, removals, claiming, and cancellation replay through the same standard splice coordinates. Ordinary removals carry outcome: 'canceled' and emit agent/inbox/discarded { message }; claiming uses pure deletions with no outcome and emits agent/inbox/claimed. Every insertion emits agent/inbox/inserted { message }. MessageId stays unique across both pending lists. Consumers that need a removed message use the claimed or discarded notification instead of depending on a pre-splice session/event view.

What a step does

Each step sends the agent's rendered system prompt, its visible tool schemas, and the session's derived history; the model's tool calls run through the guarded tool pipeline and every accepted fact is appended to the session log before the next step derives from it. Parallel-safe calls may overlap up to maxParallelToolCalls; exclusive calls run alone as ordering barriers. Cancellation is cooperative: agent.cancel() aborts the current activity and, unless keepInbox is set, clears pending work; a cancelled stream finalizes the text already delivered to the user.

---

Understand the implementation

Implementation internals — click to expand

This section explains how the package realizes the behavior above; the observable contract is covered in Use this package.

Design concept

The package is the one concrete implementation of the public Agent contract. It registers itself as the AgentFactory on ctx.agents, so consumers never import this package; ownership of each created agent lives with the caller fiber and the loop provider, converging on one memoized quiescence boundary. Every observable effect happens through session events and the agent/* taxonomy — package internals are never part of the public surface.

Request headers and adapter defaults

After agent/request, ctx.llm.prepareCall() validates adapter-owned fields and resolves reasoning-effort and output-token defaults under the active turn signal. The loop retains that exact adapter through resolution, request/header logging, and dispatch. It writes a full header for the first request, a changed envelope, an explicit message-series start, a request after surface replacement, and resume; unchanged steps, retries, and ordinary later turns in the same series inherit the latest header. Before the next waterfall, the loop removes adapter-default fields so the current route resolves them again, while explicit settings persist. An unhandled route still fails with NO_ADAPTER.

The loop deep-freezes each derived message identity on its first request and reuses that proof only within the same agent. Restored messages keep their identity; request construction does not freeze their containing event wrappers. Each request freezes its local canonical header, fresh message array, and envelope while leaving the cancellation signal live. The request-freeze decision explains ownership and measurement.

Source map

File	Role
src/index.ts	Plugin entry: AgentLoop service, config schema, declarative agent startup, factory registration
src/agent.ts	The concrete ReactLoopAgent driver: inbox, turn/step machine, cancellation
src/inbox.ts	Package-internal ReactLoopInbox: durable projection, structural commands, and loop-only claim state
src/tool-calls.ts	Tool scheduling: exclusive barriers and the bounded parallel pool
src/runtime-context.ts	Per-step runtime-context snapshot handling
src/constants.ts	DEFAULT_MAX_PARALLEL_TOOL_CALLS
src/invariant.ts	Invariant companion: request reconstruction from the session log

Creation and teardown

Creation is one rollback-covered transaction: construct a private session, concrete agent, and scoped context; await optional setup; enter both registries; announce session/created then agent/created; emit agent/session-start; only then start the driver. A setup throw, commit failure, or owner disposal rolls the transaction back without publishing either id. Teardown runs stop-and-drain, closes the session's write path, unwinds the scope, detaches the agent, then detaches the session, and every detach is bound to the exact entered object so a stale disposer cannot remove a later same-id replacement.

Persistence integration

The loop is the production acquisition point for session write handles. When ctx.sessionPersistence is mounted, create/createAgent call persistence.create(header) — storing the durable identity and taking write ownership before publication — and append the constructor seed through the handle; resume calls persistence.open(id, 'write') first (excluding a concurrent resume of the same id), reads the physically valid log through the handle, and appends interruptedTurnClosers for a log crashed mid-turn as an ordinary batch — semantic crash repair is the agent layer's job, not a storage entry point. Immediately before publication, appendUnstoredSuffix stores any events appended during the setup window (seed markers, delegation policy records), which never re-emit through session/event. Once published, the mounted backend routes the session's session/event batches, session/flush barriers, and session/disposed retirement into the active write handle by session id; the loop touches storage only through the handle it owns. The memoized teardown closes the handle — close drains any routed buffer — after the loop commits the session's closing events, provably releasing write ownership. Without a backend, sessions are memory-only and nothing else changes.

Turn and step flow

The driver owns one agent for its lifetime and runs inside ctx.agents.withInitiator(agent, ...). Its package-internal ReactLoopInbox constructor registers the standard inbox projection on the agent scope, then uses that projection for structural commands and loop-only claims. Registry reference counting keeps the shared key active until the last agent scope unloads. At a turn boundary the driver opens the durable turn, then atomically claims pending next-step input plus one queued prompt; between steps it claims only next-step input. agent/pre-step decides what enters the step. An entered decision appends its complete user/message batch before the driver can claim again, while a rejected decision appends none. Each model attempt emits one process-local start, emits every chunk only after the matching durable assistant/chunk, and emits exactly one terminal end; final assembly or message-append failure settles it as aborted, while committed follows the durable assistant/message. Each successful model call appends one message anchor citing its chunk seqs, and a cancelled stream appends an interrupted: true anchor with the delivered prefix so the next request contains what the user saw. Within a step, exclusive calls form barriers and parallel-safe calls use the bounded rolling pool; policy, durable results, and result context remain model-ordered.

Failure and cancellation

Final adapter selection, dispatch, and iteration failures arrive as terminal finishes and enter agent/request-error; a handling listener returns { kind: 'retry' } without calling next(), while an unhandled failure is terminal. Middleware, result-processing, tool, and other extension failures remain thrown and close the turn directly — plugin failure ends the turn, not the loop. Undispatched model tool calls after cancellation receive synthetic tool/call plus ABORTED_BEFORE_DISPATCH result pairs. The explicit-cancellation decision owns the signal lifecycle.

---

Further Exploration

The package-level contract is enough for most consumers; read these when you need the surrounding domain and the design rationale.

• agent package — the Agent handle, registry, and agent/* events this loop implements.
• Core subsystem — the turn flow and interception decisions.
• Session subsystem — the durable log the loop writes and derives from.
• Tools subsystem — the pipeline the loop dispatches through.
• Explicit-cancellation Agent Note — signal lifetime and cancellation races.
• Core group map — how the core packages compose.

---

Model Experience

Complete conversation request

What the model sees

For each step, the loop sends the rendered per-agent system prompt, the visible tool schemas, and the session's derived messages. It supplies provider, model, and cwd variable values but no additional fixed prose.

Token effect

System text and schemas are paid again on every step. Per-agent scoping chooses the contributions, while the authoritative assembly waterfall can alter the final request and makes its listener responsible for protocol coherence.

KV Cache effect

Append-only only while system text, schemas, and earlier history remain byte-identical under the same provider and model route. A token-bearing assembly rewrite or composition change may invalidate reuse from the first altered request token.

Retained message history

What the model sees

Accepted user messages, assistant messages, tool calls and results, injected context, and steering are logged and sent on later steps. Raw stream chunks, lifecycle boundaries, and other log-only events are excluded.

Token effect

Input grows with every surface message until a compaction replacement shadows older nodes; a multi-step tool turn resends the accumulated history each step.

KV Cache effect

Ordinary history growth is append-only and preserves reusable entries. A surface replacement or compaction invalidates reuse from the first shadowed history token.

Undispatched calls after cancellation

What the model sees

If a later request replays an aborted step, each tool call that cancellation prevented from dispatching has error code ABORTED_BEFORE_DISPATCH and result text Error: tool call aborted before dispatch.

Token effect

One fixed error result per skipped call remains in history until compaction shadows it.

KV Cache effect

Append-only; each synthetic result follows the reusable request prefix and does not invalidate existing KV-cache entries.

Known Limitations and Deferred Work

These limits define when the loop needs special care. They are current package constraints, not a task backlog.

• Classification is unary — calls whose safety depends on comparing siblings or resources must remain exclusive (rationale).
• Config labels are fresh by default — omitting sessionId creates a fresh ${id}-session-<uuid> on every startup; exact resume-or-create behavior requires an explicit stable sessionId, while resumeSessionId requires existing persisted history.
• Config agents have no per-agent persona field or setup hook — they use the deployment persona; scoped persona and tool composition are available only through the programmatic ctx.agents.create() / resume() factory options.
• No built-in turn budget — tool calls or steering continue the current turn; a policy that bounds runaway turns must cancel from an existing lifecycle extension point such as agent/turn-stopping.

Dev Note

Working context for maintainers — click to expand

None.