fix(runtime): suggest resume for an engine-known run ID only while its run directory exists (#1258)

[aviggiano]

Problem

This is a follow-up to #1264 (issue #1258). It fixes two problems in the check that refuses a run ID the workflow engine already records, before planning starts.

• The RUN_ALREADY_EXISTS message always suggests ultrafuzz resume <id>. fix(runtime): refuse a run ID the workflow engine already records before planning (#1258) #1264 mainly targets a run removed with ultrafuzz clean <id>. In that case the run directory is gone, so resume cannot work and the hint is wrong. clean leaves the engine's record behind (clean leaves task worktree registrations, ultrafuzz/<run>/* branches and the engine run record behind #1257).
• The check runs inspect --format json without --full-output, then uses ad hoc isRecord checks that accept both a bare and an {ok, data} shape. Every other inspect caller uses --format json --full-output and reads the result with commandPayload(...).

Change

• The resume hint now appears only when the run directory exists. The check looks in runsRootForProject(projectRoot), the same runs root resume uses.
  • Directory exists: ...so it cannot be launched again; choose a different --run-id, or continue it with \ultrafuzz resume ``
  • Directory is gone: ...so it cannot be launched again, even though its run directory is gone (\ultrafuzz clean` leaves the engine's record behind); choose a different --run-id`
• The check now runs inspect ultrafuzz-<id> --format json --full-output and reads commandPayload(inspected.json)?.run. The two-shape handling is removed. It does not use parseCurrentSmithersInspect, which throws on a run with no recorded history.
• Fail-open behavior is unchanged. The launch is refused only when the engine reports exactly ultrafuzz-<id>. Any other answer lets the launch go ahead, and submission still rejects a duplicate.
• No new diagnostic codes, flags or config.

Tests

• The two-shape loop is replaced by one envelope test:
  • Run directory missing (the cleaned case): the message has no ultrafuzz resume, ends with choose a different --run-id, no run directory is created, and nothing is submitted.
  • Run directory present: the message ends with the ultrafuzz resume <id> hint.
• New test: the engine reports a run whose run.id differs from ultrafuzz-<id>, and the launch goes ahead (no RUN_ALREADY_EXISTS, up is submitted).
• The existing new-run test now expects --full-output in the inspect command.
• Passed locally: the three targeted runtime tests, eslint and prettier on the changed files, pnpm typecheck for runtime, pnpm -w knip, and ESLINT_PLUGIN_DIFF_COMMIT=origin/unstable pnpm -w lint:strict:ci.

🤖 Generated with Claude Code

Confidence Score: 5/5

The PR appears safe to merge; no actionable regression was established.

Summary

The PR reads the engine’s full-output inspection envelope and suggests resume for an engine-known run ID only while its run directory exists. Tests cover missing and present directories, a mismatched engine run ID, and the updated inspection arguments.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  A[Start with run ID] --> B{Engine database exists?}
  B -- No --> P[Continue planning]
  B -- Yes --> C[Inspect with full-output]
  C --> D{Exact run ID reported?}
  D -- No --> P
  D -- Yes --> E{Run directory exists?}
  E -- Yes --> F[Refuse launch; suggest different ID or resume]
  E -- No --> G[Refuse launch; suggest different ID]

Loading

_{Reviews (1) · Last reviewed commit: "fix(runtime): suggest resume for an engi..."}