deepseek-harness/docs/tool-execution-pipeline.md

<!-- Generated by scripts/gen-doc-graphs.ts - do not edit by hand.
     Run `pnpm run gen-doc-graphs` to regenerate. -->

# Tool Execution Pipeline

This graph shows where policy, hooks, sandboxing, filesystem guards, result rewriting, final-outcome observation, and UI rendering run without changing the loop. The `tools/pre-execute` waterfall runs first, monotonic guards run next, and the `tools/execute` and `tools/post-execute` waterfalls follow; the three waterfalls may transform a call. Definition-owned `finalizeContent` and `tools/result` run afterward.

```mermaid
flowchart TD
  model["Assistant message contains tool-call block"]
  toolCall["Session event: <code>tool/call</code><br/>logged before execution"]
  presentCall["UI pending card<br/>presentCall(args)"]
  pre["<code>tools/pre-execute</code> waterfall<br/>hooks, permission, sandbox"]
  guards["Registered monotonic guards<br/>deny or abstain; identity protected"]
  denied["denied or approval refused<br/>tool body skipped"]
  approval["<code>ctx.approval</code> one-shot prompt<br/>absent or unanswerable: deny"]
  around["<code>tools/execute</code> waterfall<br/>timeout, retry, metrics (around dispatch)"]
  toolBody["Registered tool execute() body"]
  fsGate["<code>fs/write-intent</code> or <code>fs/edit-intent</code><br/>tool-fs mutations only"]
  owned["Tool-owned session events<br/><code>todo/write</code>, <code>fs/observed</code>, <code>hook/invoked</code>, <code>hook/result</code>, <code>tool/code-dispatch</code>"]
  post["<code>tools/post-execute</code> waterfall<br/>accept, block, replace, add context"]
  normalized["Registry outer normalization<br/>pipeline/result snapshot throws become isError"]
  finalize["ToolDefinition.finalizeContent<br/>last content-only invariant"]
  final["<code>tools/result</code> synchronous notification<br/>frozen authoritative outcome"]
  context["Active-batch additionalContexts FIFO<br/>injected user/message after recorded tool results"]
  toolResult["Session event: <code>tool/result</code><br/>single model-facing outcome"]
  allResults["Tool batch settled<br/>recorded tool/result events complete"]
  presentResult["UI completed card<br/>presentResult(args, result)"]
  model --> toolCall
  toolCall --> presentCall
  toolCall --> pre
  pre -->|allow| guards
  guards -->|allow| around
  guards -->|deny| denied
  guards -.->|throw| normalized
  around --> toolBody
  pre -->|deny| denied
  pre -->|ask| approval
  approval -->|allowed-once| guards
  approval -->|rejected, cancelled, unavailable| denied
  approval -.->|throw| normalized
  denied --> post
  pre -.->|throw| normalized
  toolBody --> fsGate
  fsGate --> toolBody
  toolBody --> owned
  toolBody --> around
  around --> post
  around -.->|wrapper throws| normalized
  post -.->|throw| normalized
  post --> finalize
  normalized --> finalize
  finalize --> final
  final --> toolResult
  toolResult --> presentResult
  toolResult --> allResults
  allResults --> context
```

Filesystem read-before-edit checks stay below `tool-fs` on `fs/*` events. Generic pre/post waterfalls host hooks and approval policy; `ctx.approval` resolves asks before monotonic guards, and owner policy that must not be reordered remains a registered guard. Around-dispatch concerns such as timeouts wrap `tools/execute`. The registry losslessly snapshots the candidate result and normalizes a snapshot failure before the visible definition's snapshotted `finalizeContent` callback enforces its synchronous content-only invariant. `tools/result` then observes the immutable, lossless-JSON outcome. This lets hooks span tool families without coupling the tools to one policy service. Code Mode sends both the reserved `run_code` transport and its serialized sub-calls through the pipeline; sub-calls carry the parent token, log `tool/code-dispatch`, return denials as binding rejections, and omit `additionalContexts` to preserve call/result adjacency.

Maintenance mode: curated Mermaid flow; exact tool schemas and event signatures live in generated catalogs.