fix(zcode): keep a team-configured hook timeout instead of the per-event default

[Dev-next-gen]

Summary

toZcodeEntry reads its timeoutMs only from the ZCODE_TIMEOUT_MS table introduced in #555 and never looks at def.timeout, so on ZCode a timeout the team wrote down is silently dropped: a timeout: on a hook in hooks/hooks.yaml, and the builtin.overrides.<key>.timeout knob the usage guide documents ("Only the timeout of a built-in hook can be overridden"). Before #555 the writer honored it (timeoutMs: def.timeout * 1000). The per-event table is the right default for the builtin dispatch hooks — a session-start pull really does need more than the 10–15s shell defaults — but it also took over the values a team had stated explicitly.

I kept the table as the default and let a stated timeout win, which is what every other tool's writer already does.

Type of Change

• Bug fix (non-breaking change that fixes an issue)
• New feature (non-breaking change that adds functionality)
• Breaking change (fix or feature causing existing behavior to change)
• Documentation only
• Refactor / internal cleanup

Test Plan

• npx tsc --noEmit passes
• npx vitest run passes (see scope note below)
• Added/updated tests for the change

End-to-end with the built CLI, single-repo mode, --agent zcode, .teamai/hooks/hooks.yaml:

hooks:
  - id: slow-sync
    description: nightly sync that needs five minutes
    event: Stop
    command: my-slow-sync
    timeout: 300
builtin:
  overrides:
    Hook dispatch stop:
      timeout: 240

After npm run build and teamai hooks inject, the Stop entries in .zcode/cli/config.json:

before   teamai hook-dispatch stop --tool zcode     timeoutMs = 60000
         my-slow-sync                               timeoutMs = 60000

after    teamai hook-dispatch stop --tool zcode     timeoutMs = 240000
         my-slow-sync                               timeoutMs = 300000

Unit side, src/__tests__/zcode.test.ts gains one case driving reconcileHooks(..., 'zcode', ...) with both fields. On the current main it fails with expected 60000 to be 300000; with the change the whole ZCode file passes (11/11).

I also ran the neighbouring hook suites — hooks-golden, hooks-team, hooks-dispatch-format, builtin-hooks, hooks-reconcile-scope, hooks-hasteamaihooks, hooks-security, hooks-shell-check — 72/72 green. The golden fixtures matter here: builtinHookDefs sets withTimeout = false for zcode, so builtin defs carry no timeout and the rendered output for a config without overrides is byte-identical. Only a config that states a timeout changes.

Scope note, so you can weigh it: I ran the suite on Windows, where a number of pre-existing cases fail for reasons unrelated to this change (POSIX path separators asserted in test strings, 0600 mode checks, git-backed fixtures timing out). The eight suites above and the full ZCode file are green here, and none of the touched code is platform-conditional beyond the win32 branch #555 added, which this change leaves alone.

Related Issues

Follow-up to #555, which added the per-event table.

Notes for Reviewers

The one judgement call is precedence. I read the table as ZCode's default rather than an override, because HookDef.timeout is documented as "per-hook timeout in seconds (tool-specific; omitted = tool default)" and the guide lists builtin.overrides.<key>.timeout as a supported knob. If you would rather the table stay a hard floor for the builtin dispatch hooks and only team hooks from hooks.yaml be honored, that is a one-line change in the same spot and I am happy to make it.

I did not add a CHANGELOG entry: most recent fixes here don't carry one, so it looked like something you curate at release time. Say the word and I will add it.

Found by a defect-hunting pipeline I build and run (Dev-next-gen), using Claude Code with Anthropic's Claude Opus 5.