deepseek-harness/docs/architecture.md

DeepSeek Harness Architecture

The DeepSeek Harness SDK builds agent harnesses on Cordis. The principle is simple: everything is a plugin. The shipped loop is one plugin, not a privileged kernel.

Overview

A harness is one Cordis context. Packages contribute service keys, typed events, and disposable registrations: services expose stable calls (ctx.llm, ctx.tools, ctx.sessions), events provide interception and notifications (agent/request, tools/pre-execute, session/event), and registrations install prompt sections, tools, providers, adapters, or listeners.

packages/core/ groups the default agent flow; surrounding capabilities are equally first-class Cordis plugins.

Default Services

ctx key	Package	Role
—	dsh-scope	scoped-context registration primitive (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 agent registry, public Agent handle, agent/* events
ctx.agentLoop	dsh-agent-loop	shipped ReactLoopAgent driver

Capability Services

ctx key	Package family	Role
ctx.llm	llm/	adapter registry and streaming model calls
ctx.bash	bash/	foreground/background command execution
ctx.sandbox	sandbox/	same-world process confinement (argv wrapping, per-call policy)
ctx.codeRuntime	code-runtime/	model-written program execution
ctx.fs	fs/	filesystem provider primitives and policy events
ctx.skills	skill/	skill provider registry and progressive disclosure
ctx.web	web/	search/fetch provider registries
ctx.compact	compact/	session-log compaction
ctx.subagents	subagent/	named delegation providers
ctx.tasks	tasks/	background task registry + generic task_* control tools
ctx.workflows	workflow/	script-driven multi-agent orchestration
ctx.sessionPersistence	session-persistence/	durable storage for session logs

Event

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

Event Domains

• Session events are durable, replayable facts. Turn and step boundaries, user input, assistant output, tool calls, tool results, steering, compaction records, and tool-owned durable facts append to the session log and flow through session/event.
• Agent events carry the live Agent handle for status, diagnostics, prompt admission, call-config shaping, result validation, and continuation policy.
• Capability events belong to the seam that owns the action. tools/*, llm/*, system-prompt/*, fs/*, and subagent/* let policy and adapters attach 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 drains work, assembles requests, streams model answers, executes tools, applies continuation policy, and checkpoints state. Every pause is a service call or event available to plugins.

A session is one agent's append-only event log. A turn drains one queued batch and runs until the model stops asking for tools and no plugin requests continuation. A step is one model request plus the tool executions caused by that response. In the flow below (sequence companion), quoted names are durable session events and event names are extension points.

Turn Flow

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 queued messages
  emit agent/status(running)
  TURN:
    'turn/start'
    each queued message -> agent/prompt-submit
      allowed prompt -> 'user/message' plus injected context
    every prompt blocked -> 'turn/end'(rejected)
    STEP loop:
      drain steering
      assemble system prompt and tool schemas
      agent/session-prefix (first step)
      agent/pre-step
      'step/start'
      snapshot the derived messages (the reconstruction boundary)
      agent/request (config only) -> log request/header -> llm/stream (frozen)
        'assistant/chunk'
      agent/step-result
      'assistant/message'
      each tool call:
        'tool/call'
        tools/pre-execute -> monotonic guards -> tools/execute -> tools/post-execute -> tools/result
        'tool/result'
      append post-tool context and steering
      '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

Prompt assembly is single-path: the loop sends renderPrompt(await assemble(assembleContextFor(agent))); the helper couples the explicit agent and scope. Plugins contribute ordered sections, tool schemas, and named variables interpolated as {{name}} at render — strictly, so an unknown or valueless reference fails the turn instead of shipping a hole. dsh-system-prompt owns the openers — the static harness:identity section (order −100) and the deployment's global default persona (order 0, shadowable by a same-named agent-scoped section) — while the loop registers the model/cwd variables; prompt-fact ownership is pinned by the prompt-variables RFC.

Post-tool context lands after all tool results so tool-call/result adjacency stays stable. Steering drains between steps; ordinary leftover steering after a turn is re-queued as input. A terminal agent/turn-stop is the explicit exception: it runs after ordinary continuation and steering folding, then remains authoritative through turn close and flush so steering from those later listeners is discarded rather than becoming another step or turn; ordinary queued prompts are preserved.

Failure Boundaries

The turn is the containment boundary. A throwing listener, adapter error finish, or failed step ends the current turn with an error reason and reports live diagnostics through agent/error; it does not kill the driver loop. cancel() clears queued and steering work, aborts the active model/tool boundary when possible, and records the appropriate turn end. Disposal stops the loop, awaits quiescence, unregisters the agent, and lets service disposers drain.

Every session event is turn-enclosed. Reloading a crashed session preserves the interrupted tail and closes it with a synthetic interrupted turn end. A failure after the durable turn has closed reports through agent/error only because no safe in-turn position remains. A turn ends with one TurnEndReason (completed, aborted, error, disposed, max-tokens, rejected, or interrupted); per-variant semantics are in session.md § TurnEndReasonMap.

Agent Handles

ctx.agents owns live agents and returns AgentHandle { agent, dispose() }. Plugins drive Agent through send(), steer(), inject(), cancel(), and whenIdle(). The caller fiber and factory provider structurally co-own programmatic lifecycles; the consumer handle is the only other teardown capability, and all owners await one disposer.

Agent Scope

Every live agent owns agent.ctx (dsh-scope). Agent-local registrations shadow globals and unwind with the agent; async effects such as background-task cleanup are awaited. Scoped listeners hear only that agent through an opaque routing carrier. CreateAgentOptions.setup(agentCtx) composes this world before publication. Dev invariants verify carrier/subject alignment. Rationale: agent-scope RFC; subagent composition controls are a separate feature.

State

Session Log

The session log is the source of truth. deriveMessages() projects session events into the Message[] sent to the model; raw assistant/chunk events stay in the log for replay and UI fidelity. Replay, fork, resume, transcript rendering, telemetry, and persistence all derive from the same event stream.

Model-visible ⟺ logged: the log reconstructs every request — messages at step/start fronted by the header's session prefix, headers by folding request/header — and dev invariants assert this (reconstructability RFC).

Durability is a plugin concern. Persistence backends buffer synchronous session/event notifications and the loop awaits a turn-end checkpoint before moving on. The SessionPersistence seam stores SessionEvent directly, with metadata in SessionHeader; JSONL and SQLite share one contract suite.

Model Content

Messages are arrays of typed content blocks (text, reasoning, tool-call, tool-result). The union derives from the merge-extensible ContentBlockMap; the same pattern types MessageSource, FinishReason, TurnTrigger, and TurnEndReason. New block types are coordinated across adapters, UI bridges, compaction pricing, and persistence, so block types remain a repo-wide contract.

Streaming is a raw chunk protocol (block-start through finish) with BlockAssembler as the shared chunk-to-block assembler. The loop logs raw chunks while assembling them for dispatch. LlmAdapter is the provider seam: subclass, implement stream(), and register with ctx.llm.registerAdapter(models, adapter). StreamChunk conventions live in llm-streaming.md.

Extension And Composition

Capability Pattern

A swappable capability usually splits into interface / implementation / consumer: the interface owns its ctx key and events, an implementation registers a backend, and a consumer exposes model behavior through tools or prompts. Bash is the reference; the capability graph shows every family.

Some seams bend the template deliberately. LLM keeps interface and consumer vocabulary together because adapters are the implementations. Filesystem adds policy gates around provider primitives. Web is one service with search and fetch provider registries, so provider swaps do not rename model tools. Skills and subagents use named provider registries; local skills scan project/user roots, and other providers can add embedded or remote catalogs without registry/tool changes. Subagents spawn fresh, fork from the parent's completed-turn prefix, or use ACP children (subagent.md).

Bundles And Apps

dsh-agent-core is the default composition bundle: one plugin loading the shared spine (README). App packages compose it with a front door and boot bin: dsh-stdio-agent for terminal REPL, and dsh-acp-agent for ACP over JSON-RPC stdio with no stdout logger (ui/). A deployment is a thin cordis.yml leaf: swappable backends, one app entry, and optional product tools (examples/, runnable wirings, graph atlas).

Where New Behavior Goes

New behavior should attach to a documented extension point; changing the shipped loop requires updating this map.

Goal	Mechanism
Add a model provider	register an adapter on ctx.llm
Add a model-facing capability	register a tool on ctx.tools; schemas flow into prompt assembly
Add command execution	implement and register a ctx.bash backend
Add a long-running/background capability	register the work on ctx.tasks; the generic task_* tools collect/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 prompts, requests, tool use, or continuation	listen on the relevant agent/* or tools/* waterfall; use serial agent/turn-stop for a monotonic terminal stop
Add a session-stable request prefix outside history	compose it on agent/session-prefix, once per loop instance; logged on the request header
Add UI or editor integration	drive ctx.agents and render from session/event
Add durable session state	add a SessionEventMap member and render/replay from the log
Fork a live session	use ctx.sessions.fork(source, boundary?, childSessionId?)
Scope a tool, prompt section, or listener to ONE agent	register it through 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 event and service signatures in events
• services catalogs
• package contracts in the package map
• RFCs