deepseek-harness/docs/architecture.md

DeepSeek Harness Architecture

English | 中文

DeepSeek Harness SDK uses Cordis: everything is a plugin, including the loop.

Overview

Harnesses are Cordis contexts with package-contributed services, typed events, and disposable registrations.

packages/core/ groups the default agent flow; capabilities remain plugins.

Default Services

ctx key	Package	Role
—	dsh-scope	scoped-context registration and shared layer storage (library)
ctx.sessions	dsh-session	in-memory event-sourced sessions
ctx.systemPrompt	dsh-system-prompt	ordered prompt sections, tool schemas, and prompt variables
ctx.tools	dsh-tools	tool registry and execution pipeline
ctx.agents	dsh-agent	live agents, delegated creation, agent/* events, and process-local initiator scope
ctx.agentLoop	dsh-agent-loop	concrete Agent driver

Capability Services

ctx key	Package family	Role
ctx.llm	llm/	adapter registry and streaming model calls
ctx.tokenMeter	llm/token-meter	singleton replay-aware request/surface pressure
ctx.bash	bash/	foreground/background command execution
ctx.processes	process/	managed child-process groups under the bash executors
ctx.pty	pty/	owner-scoped persistent terminal sessions
ctx.sandbox	sandbox/	same-world process confinement (argv wrapping, per-call policy)
ctx.sandboxPolicy	sandbox/	shared sandbox policy home
ctx.codeRuntime	code-runtime/	model-written program execution
ctx.fs	fs/	filesystem provider primitives and policy events
ctx.lsp	lsp/	semantic navigation registry
ctx.skills	skill/	skill provider registry and progressive disclosure
ctx.web	web/	search/fetch provider registries
ctx.compact, ctx.toolResultPrune	compact//compact-tool-result-prune	summary compaction; optional model-free result pruning
ctx.subagents	subagent/	named delegation providers
ctx.planMode	plan/	logged plan collaboration state
ctx.tasks	tasks/	background task registry + generic task_* control tools
ctx.workflows	workflow/	script-driven multi-agent orchestration
ctx.goals	goal/	persisted same-session goals
ctx.sessionPersistence	session-persistence/	durable session-log storage
ctx.sessionQuery	session-query/	Live-preferred exact/filter/trace interface, SQLite FTS backend, and workspace-authorized model tools
ctx.sessionTitle	session-title/	log-backed fallbacks plus one optional asynchronous provider
ctx.invariants	support/invariants	package-name-selected registry for package-owned runtime checks

Event

Events form the service extension API; see the catalog and producer/consumer map.

Event Domains

• Session events are durable facts appended to the log and emitted through session/event.
• Agent events carry the live Agent for status, prompt admission, request shaping, validation, and continuation.
• Capability events let owning seams attach policy and adapters without importing the loop.

Interception Semantics

Waterfall events behave like around-middleware: a listener delegates by calling next(); returning without it vetoes or takes over. Full rule: Cordis waterfall semantics.

Default Loop Lifecycle

The loop runs through plugin services and events.

A session is append-only. Each ordinary turn claims one queued message; injection claims none. Successors await the preceding checkpoint but may share its running interval (decision). A step is one model request plus tools; quotes in the sequence below mark durable events.

Creation without an id mints <config-id>-session-<uuid>; sessionId resumes or creates, while resumeSessionId requires history. Resume restores lineage and delegation depth before publication. Setup failures emit agent-loop/config-start-failed; teardown is silent.

Turn Flow

choose declarative identity and fresh/resume path
  -> prepare private session + agent.ctx -> await unpublished setup
  -> enter session + agent -> session/created -> agent/created
  -> enable driving -> agent/session-start(source) -> start driver
forever:
  wait for a queued message
  emit agent/status(running)
  TURN:
    'turn/start'
    claimed message + contexts -> agent/prompt-submit
      allowed prompt -> 'user/message' with prompt-prefix context baked in; append separate contexts
      blocked prompt -> 'prompt/blocked' -> 'turn/end'(rejected)
    STEP loop:
      drain steering with the same prefix/separate context placement (no prompt-submit)
      assemble system prompt and tool schemas
      agent/session-prefix (first step)
      agent/pre-step
      snapshot the derived messages (the reconstruction boundary)
      'step/start'
      agent/request (config only) -> log request/header -> checkpoint -> llm/stream (frozen)
      on final adapter-path or terminal in-band failure:
        'step/end'
        agent/request-error(original error, failure facts, immutable prior failures, signal)
        retry in the next numbered step or preserve the original error
      otherwise:
        'assistant/chunk'
        agent/step-result
        'assistant/message' (transformed content or empty success anchor after step-result rejection)
        schedule tool calls by ctx.tools.executionMode:
          exclusive -> one-call barrier
          parallel -> rolling pool, <= maxParallelToolCalls in flight; reclassify before start
          each start -> 'tool/call' -> ordered tools/pre-execute -> checkpoint -> concurrent tools/execute
          each model-order result -> ordered tools/post-execute -> 'tool/result'
        append accepted tool-batch context after all recorded results, then steering
        agent/post-step -> checkpoint complete response/results
        'step/end'
        agent/turn-continuation
        agent/turn-stop (terminal policy)
        stop unless tools or continuation policy ask for another step
    'turn/end'
    checkpoint persistence and notify idle/running status

Steps assemble ordered prompt sections, tool schemas, and variables; unknown references fail turns. dsh-system-prompt owns identity and persona; the loop supplies model and cwd (ownership).

Async inject() and post-tool additionalContexts settle after results; steering drains before agent/post-step. Leftovers queue. Terminal agent/turn-stop remains authoritative through close/flush and discards later steering, not queued prompts.

Pruning precedes summaries; overflow retries require durable progress. Bounded retries compose on agent/request-error; cancellation wins (compaction, retry).

Failure Boundaries

Adapter failures close the step before agent/request-error with exact Error, LlmFailure, and history. Retries open steps; success clears history; exhaustion stores failure on turn/end. Failed chunks commit nothing.

Other failures use agent/error. Cancellation and disposal beat recovery; undispatched tools get synthetic tool/call/ABORTED_BEFORE_DISPATCH pairs. The signal retires before turn/end. Effective cancel() emits its cause, clears queues, and aborts; observers cannot veto, idle calls emit nothing, and durability records aborted. Disposal awaits quiescence (decision).

Session events are turn-enclosed; reload closes an interrupted tail with a synthetic interrupted turn end. Post-close failures use agent/error. Each turn has one TurnEndReason.

Agent Handles

ctx.agents returns AgentHandle { agent, dispose() }. Plugins use intent helpers followup(), queue(), steer(), and inject(); callers with exact routing facts use mandatory-field send() (decision). cancel() and whenIdle() control lifecycle. Caller, provider, and handle co-own teardown.

Agent Scope

Each agent owns a scoped agent.ctx over global tool, prompt, and command storage (decision); scoped listeners filter and contributions unwind with awaited cleanup. CreateAgentOptions.setup(agentCtx) composes before publication; typed resolvers derive carrier checks from Events and scopeTarget (gates). AgentLoop runs inside ctx.agents.withInitiator(); private orchestration derives agent.session, while turn, step, signal, cwd, and authority stay explicit (decision). See agent scope and subagent composition.

State

Session Log

The session log is authoritative. deriveMessages() projects model history; raw assistant/chunk events preserve replay and UI fidelity. Fork, resume, transcripts, telemetry, and persistence share that stream.

Model-visible ⟺ logged: step/start messages plus the header's session prefix and folded request/header reconstruct every request; dsh-agent-loop/invariant asserts this through ctx.invariants (decision).

Durability is a plugin concern; backends buffer synchronous session/event notifications. Checkpoints drain before adapter dispatch, recorded top-level tool calls before tool dispatch, complete response/result batches at agent/post-step, and final turn ends. SessionPersistence stores SessionEvent plus SessionHeader metadata; JSONL defaults to checksummed Zstandard, with SQLite under one contract (decision).

ctx.sessions.appendOutOfBand() joins plugin-owned log-only events to an open turn or creates a balanced, flushed zero-step turn. session/title folds latest-wins with source seqs and provenance; its immediate fallback and sole optional async provider never delay the agent response. Forks inherit titles (decision).

Model Content

Messages use typed blocks from merge-extensible ContentBlockMap; the same pattern types MessageSource, FinishReason, TurnTrigger, and TurnEndReason. New blocks coordinate adapters, UI, compaction, token metering, and persistence; replay measurements live in token-meter.md.

Streaming uses raw chunks and BlockAssembler. Each LlmAdapter.stream() is one provider attempt; adapters report facts and agent/request-error owns recovery. The loop logs chunks and successful provenance/replay state. Remote adapters use per-read idle watchdogs. Replay state crosses routes only when they share an adapter instance (contract).

Extension And Composition

Capability Pattern

A swappable capability usually splits into interface / implementation / consumer: service/events, a backend, and model-facing tools/prompts. Bash is the reference; the capability graph maps each family.

Exceptions combine layers: LLM interface/consumer; filesystem policy; web registries; named skill/subagent providers. Subagents spawn fresh, fork a completed-turn prefix, or use ACP children (subagent.md).

dsh-workspace-context composes baselines on agent/session-prefix and appends ctx.fs-discovered nested changes on tools/post-execute; its decision records isolation. dsh-paths owns shared paths.

Bundles And Apps

dsh-agent-spine-demo bundles a spine and optional goals. App packages own TUI, CLI, ACP automation, and JSON-RPC front doors (README, acp/, ui/). dsh-jsonrpc-agent boots external cordis.yml; the Python SDK supplies a default only without explicit config (Python SDK). Thin deployments use swappable backends and optional tools (examples/, runnable wirings, graph atlas).

Where New Behavior Goes

New behavior attaches to a documented extension point; a loop change updates this map.

Goal	Mechanism
Add a model provider	register an adapter on ctx.llm
Add a model-facing capability	register on ctx.tools; schemas enter prompt assembly
Add shell execution	implement and register a ctx.bash backend (the local one spawns through ctx.processes)
Add persistent terminal execution	register a ctx.pty backend and dsh-tool-pty
Add a human command	register on ctx.commands; adapters discover and dispatch it without a model turn
Add background work	register on ctx.tasks; generic task_* tools collect or stop it
Add filesystem access or policy	implement a ctx.fs provider or listen on fs/* policy events
Confine spawned processes	a ctx.sandbox backend; consumers wrap their argv before spawning
Intercept a request, tool, or turn	use its agent/* or tools/* event; agent/turn-stop is the serial terminal stop
Add a session-stable prefix outside history	compose agent/session-prefix; the request header logs it
Add UI or editor integration	drive ctx.agents and render from session/event; terminal-only overlays use ctx.tui
Add durable session state	add a SessionEventMap member and render/replay from the log
Add asynchronous session-title generation	register the sole provider on ctx.sessionTitle
Manage a same-session objective	use ctx.goals; continue through Agent and agent/*
Fork a live session	use ctx.sessions.fork(source, boundary?, childSessionId?)
Scope a registration to one agent	use that agent's agent.ctx (see Agent Scope)

The extension cookbook carries plugin skeletons and the feature-to-seam map; step-by-step guides cover packages, tools, LLM adapters, and vendored packages.

Quick Reference

• Domain terms in the glossary
• Type definitions in core-data-structures/
• Exact signatures in the event and service catalogs
• package contracts in the package map
• Agent Notes