461c5ac148
- Change default base branch from origin/integration to main - Add --noFetch flag to skip git fetch (for offline/sandbox use) - Add try/except with rollback on spawn failure - Improve error message for missing review-gate - Add Codex auth.json symlink to use-skills.sh - Include worker orchestration AAR from 2026-01-13 Addresses pain points from worker-orchestration-aar-2026-01-13.md Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
57 lines
2.4 KiB
Markdown
57 lines
2.4 KiB
Markdown
# Worker Orchestration AAR (2026-01-13)
|
|
|
|
## Goal
|
|
Parallelize three bd issues using `worker spawn` and background worker agents.
|
|
|
|
## Scope
|
|
- Issues: `talu-gsga`, `talu-jid5`, `talu-w8oq`
|
|
- Base branch: `main`
|
|
- Worker model: `sonnet-4.5`
|
|
|
|
## Environment
|
|
- Repo: `/home/dan/proj/talu`
|
|
- Network: restricted sandbox (fetch requires escalation)
|
|
- Tools: `worker`, `bd`, `git`
|
|
|
|
## What Happened
|
|
1. Ran `worker spawn` using positional args; command failed due to CLI syntax.
|
|
2. Retried with `-t` and `-d` flags; `worker` attempted `git fetch` and failed due to network restriction.
|
|
3. Retried with `-f main`; fetch still attempted and failed.
|
|
4. Partial worktrees and branches were created without worker registry entries.
|
|
5. Manually removed worktrees and deleted branches.
|
|
6. Re-ran `worker spawn` with network escalation; workers created successfully.
|
|
7. `review-gate` was not found, so review integration was disabled.
|
|
8. Rendered worker prompts and launched background workers.
|
|
|
|
## What Went Well
|
|
- After network access, `worker spawn` created worktrees/branches reliably.
|
|
- Prompt rendering and background worker launch were straightforward.
|
|
|
|
## Pain Points
|
|
- `worker spawn` always attempts `git fetch`, even when `--fromBranch` is local.
|
|
- Default base branch is `origin/integration`, which is not present in this repo.
|
|
- Spawn failures left behind branches and worktrees without worker registry state.
|
|
- Missing `review-gate` produces warnings without guidance on setup.
|
|
- Network access requirements are easy to miss during first-time use.
|
|
|
|
## Impact
|
|
- Time lost to retries and cleanup before workers could start.
|
|
- Non-obvious failure modes and manual recovery steps.
|
|
|
|
## Observed Errors
|
|
- `spawn does not expect non-option arguments at "talu-gsga"`
|
|
- `fatal: not a valid object name: 'origin/integration'`
|
|
- `ssh: connect to host 192.168.1.108 port 2222: failure`
|
|
- `WARN: enableReview: failed for <id>: review-gate not found`
|
|
|
|
## Recommendations
|
|
1. Allow `worker spawn` to skip `git fetch` when the base branch is local.
|
|
2. Make the default base branch configurable or auto-detect a local main branch.
|
|
3. Roll back branch/worktree on spawn failure to avoid manual cleanup.
|
|
4. Improve error messaging to distinguish network vs branch-not-found.
|
|
5. Provide setup guidance when `review-gate` is missing.
|
|
|
|
## Questions for Worker Team
|
|
- Can `worker spawn` be configured to avoid network fetches?
|
|
- Is there a way to set a global default base branch (e.g., `main`)?
|