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 whose packages contribute 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.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 exhaustive events 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 shipped loop runs prompt-to-checkpoint work through plugin services and events.

A session is append-only. Each ordinary turn claims one queued send() item; injection claims none. A successor awaits the preceding claimed turn's checkpoint but may share its running interval (decision). A turn ends when model and plugins stop it; a step is one model request plus tools. In the sequence below, quotes mark durable events.

Without an id, creation 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
  claimed message -> agent/prompt-submit
    blocked prompt -> park without opening a turn
    allowed prompt:
      emit agent/status(running)
      'turn/start'
      append prompt + additional contexts as separate 'user/message' events
    STEP loop:
      agent/step
      drain injected context and steering (steering bypasses prompt-submit)
      assemble system prompt and tool schemas
      snapshot the derived messages (the reconstruction boundary)
      'step/start'
      agent/request (config only) -> log request/header -> llm/stream (frozen)
      'assistant/chunk'
      'assistant/message'
      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 -> concurrent tools/execute
        each model-order result -> ordered tools/post-execute -> 'tool/result'
      drain accepted tool context and steering
      'step/end'
      continue for tools or steering unless a result concluded the turn
      otherwise agent/stopping -> drain -> continue only for steering
    'turn/end' -> agent/idle
  start the next waking queued message, or emit agent/status(idle)

idle inject:
  append 'user/message'
  do not open a turn or run the model

Each step assembles ordered prompt sections, tool schemas, and variables; unknown references fail the turn. dsh-system-prompt owns identity and persona, while the loop supplies provider, model, and cwd (prompt ownership).

Tool-time context—including active-turn inject() and post-tool additionalContexts—settles after results. Steering drains at that boundary and requests another step. Idle inject() appends context immediately without changing turn numbering; persistence drains it eagerly.

Pruning precedes summaries; overflow retries require durable progress. Recovery runs through agent/request-error between the failed step and turn closes. A handling policy calls agent.retry() to schedule one retry turn; cancellation wins (compaction, retry).

Failure Boundaries

Adapter failures close the step before agent/request-error receives the exact Error, normalized LlmFailure, and turn signal. A handling listener calls agent.retry(); the loop closes the failed turn and opens another from durable history without an idle notification. Exhaustion leaves the failed turn/end terminal. Failed chunks commit no message or tool call.

Other failures use agent/error. Cancellation and disposal beat recovery; undispatched tools get synthetic tool/call/ABORTED_BEFORE_DISPATCH pairs. Effective cancel(cause) emits its cause before clearing queues and aborting; observers cannot veto and idle calls emit nothing. Durability records aborted for user or parent cancellation and disposed for teardown, which awaits quiescence. The cause changes reporting, not late result-context handling (decision).

Turn and step events are turn-enclosed; idle injected user/message events may sit between turns. Reload closes an interrupted tail with a synthetic turn end. Post-close failures use only agent/error. Each turn has one TurnEndReason.

Agent Handles

ctx.agents owns live agents and returns AgentHandle { agent, dispose() }. Plugins use complete send() options or the followup(), steer(), and inject() presets; cancel() and whenIdle() control lifecycle. One awaited disposer coordinates teardown ownership.

Agent Scope

Each agent owns a scoped agent.ctx; shared storage overlays global tool, prompt, and command entries while preserving domain views (decision). Scoped listeners filter dispatch, and every scoped contribution unwinds with awaited cleanup. CreateAgentOptions.setup(agentCtx) composes before publication. Typed resolvers derive carrier checks from merged Events and scopeTarget (semantic gates). See agent scope and subagent composition. AgentLoop runs inside ctx.agents.withInitiator(); private orchestration derives agent.session, while turn, step, signal, cwd, and authority remain explicit (decision).

State

Session Log

The session log is authoritative. deriveMessages() projects model history; raw assistant/chunk events remain for replay and UI fidelity. Fork, resume, transcript rendering, telemetry, and persistence derive from the same stream.

Model-visible ⟺ logged: the log reconstructs every request from the messages at step/start and the folded request/header; the package-owned dsh-agent-loop/invariant can assert it through ctx.invariants (reconstructability).

Durability is a plugin concern. Backends eagerly drain synchronous session/event notifications. The semantic checkpoint policy uses session/flush as an observation barrier before adapter dispatch, before top-level tool dispatch, and at agent/step before the next request. SessionPersistence stores SessionEvent directly and metadata in SessionHeader; 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 normalized failure facts and a handling agent/request-error plugin calls agent.retry(). 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 injects the baseline at the first agent/step and appends ctx.fs-discovered changes through 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
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/stopping is the stop boundary
Add model-facing context	call agent.inject(); it appends a sourced user/message without creating a turn
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