Consolidate skills into workon-flow and workon-delivery#2
Conversation
There was a problem hiding this comment.
Review findings at time of initial review. See inline comments on specific files for resolution status.
Critical
- Command duplication —
context,cleanup,worktreein both skills - Circular dependency — skills reference each other without documenting the pattern
- Missing error handling guidance — no guidance when CLI commands fail
- Test failure logic contradiction — mandatory gate vs stop condition overlap
Important
- Ambiguous auto mode for cross-repo — unclear if subtasks can run in auto mode
- Dangerous
--yesin auto loops — no test verification before pushing
-agent:code-reviewer:claude-opus-4-6
Replace 8 per-command skill files with 2 consolidated skills that cover the full workflow. Both skills include dynamic context injection for session state, detailed workflow documentation, and auto mode support for minimal-intervention execution on well-validated tickets. workon-flow: ticket discovery, validation, cross-repo orchestration workon-delivery: PR lifecycle, CI monitoring, review handling, merge Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…eview Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add worktree primitives so tickets can be offloaded to isolated directories where background Claude agents run the full auto-mode pipeline, while the main worktree stays free for interactive use. New commands: - workon worktree add <id>: create worktree + branch from base - workon worktree list [--json]: show all worktrees with pipeline status - workon worktree remove <id> [--keep]: remove worktree + branch - workon worktree status [id]: detailed status view - workon worktree update-status --stage <stage>: agent progress reporting - workon start <id> --worktree: sugar for worktree add Integration: - cleanup: detects linked worktree, removes instead of branch delete - context: shows "Background Work" section, flags blocked worktrees - .gitignore: idempotently adds .worktrees/ and .workon-status.json Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Remove command duplication (cleanup from flow, context from delivery) - Add companion skills note documenting intentional handoff pattern - Add error handling guidance to both skills - Clarify test failure logic (modified vs unmodified code) - Clarify auto mode eligibility for cross-repo subtasks - Add local test verification before --yes pushes in auto mode Co-Authored-By: agent(claude-opus-4-6) <agent@noreply>
386722b to
963ac96
Compare
| | `workon pr-push [pr-number] --yes` | Push changes + re-request review | | ||
| | `workon pr-ready [pr-number] --yes` | Mark draft PR as ready for review | | ||
| | `workon ci-status [branch]` | Check CircleCI status | | ||
| | `workon ci-failure [job-number]` | Get detailed CI failure output | |
There was a problem hiding this comment.
Cleanup command kept here in delivery (removed from flow). Context command removed from here (kept in flow). Worktree commands kept in flow only.
|
|
||
| # Workon Flow | ||
|
|
||
| > **Note**: workon-flow and workon-delivery are companion skills. Flow handles ticket discovery through implementation; Delivery handles the PR lifecycle through merge. They hand off to each other at defined boundaries. |
There was a problem hiding this comment.
Added companion skills note to document the intentional handoff pattern between flow and delivery.
|
|
||
| See `commands-reference.md` for detailed flag reference. | ||
|
|
||
| ## Error Handling |
There was a problem hiding this comment.
Added error handling guidance: retry once on network errors, report immediately on invalid IDs, never proceed if a step fails.
| - **Max 3 CI fix attempts** — after 3 failures, stop and report to user | ||
| - **Review changes > 20 lines of rework** — if a reviewer requests large changes, stop and consult user | ||
| - **Scope creep detection** — if implementation requires changes outside what the ticket describes, stop | ||
| - **Test failures in unmodified code** — if tests you didn't write or modify are failing (flaky tests, environment issues), stop and report to user |
There was a problem hiding this comment.
Clarified: this applies to tests in unmodified code only (flaky/environment). The mandatory test gate in step 4 handles failures in modified code.
| - `ready: true` from deep validation | ||
| - No semantic findings with category `coverage-gap`, `contract-mismatch`, or `thin-subtask` | ||
| - No subtasks (single-repo ticket) — parent-level cross-repo orchestration requires guided mode | ||
| - Individual subtasks within a cross-repo ticket CAN run in auto mode if they independently pass auto-eligibility checks |
There was a problem hiding this comment.
Clarified: parent-level cross-repo orchestration requires guided mode, but individual subtasks CAN run in auto mode independently.
| In auto mode, CI fix is a loop with a retry limit: | ||
|
|
||
| 1. `workon ci-status` — if green, move on | ||
| 2. If red: `workon ci-failure` → diagnose → fix → **run tests locally to verify fix** → `workon pr-push --yes` |
There was a problem hiding this comment.
Added local test verification step before each --yes push in auto mode (both CI fix and review feedback loops).
Summary
!blocks) for automatic session stateDepends on
Test plan
~/.claude/skills/workon-flow→ reposkills/workon-flow/and verify skill discovery~/.claude/skills/workon-delivery→ reposkills/workon-delivery/and verify skill discovery.github/PULL_REQUEST_TEMPLATE.mdworkon validate <id> --deep --json→ automatability report🤖 Generated with Claude Code