fix: improve pr-create and pr-status reliability, consolidate pr-merge#13
Conversation
pr-create: Add existing-PR check using `gh pr list` (exits 0 when no PR exists) instead of `gh pr view` (exits non-zero). Make confirmation step resilient with the same approach. pr-status: Rewrite from snapshot-only to full readiness checker — watches CI to completion, polls for review bot comment with timestamp validation against the current CI run, scans for blocker keywords, and reports a clear READY/NOT READY verdict. pr-merge: Remove duplicated CI watching, comment polling, and blocker scanning (Steps 3-6). Delegate readiness checking to /pr-status, keeping only merge-specific logic: local verification, review feedback assessment, fix loop, merge execution, and return to main. Fixes #12 https://claude.ai/code/session_01Ch89tRMukrCMshPhYskRnC
…status - Delete pr-create.md: redundant with Claude Code's built-in PR creation and GitHub PR templates - Delete pr-merge.md: merge is a single gh command once status says ready - Rewrite pr-status.md as the single readiness checker: watches CI to completion, polls for the review comment with timestamp validation against the current CI run, scans for blockers, and reports a clear READY TO MERGE / CHANGES NEEDED verdict - Update agent-team-development and subagent-driven-development skills to use /pr-status in a fix loop (check → fix → push → re-check) instead of invoking /pr-merge. Agent decides when to run staff-code-reviewer on fixes. - Update context-recovery.md, git-guard.py, and CLAUDE.md to remove all references to deleted commands Fixes #12 https://claude.ai/code/session_01Ch89tRMukrCMshPhYskRnC
…t into plan-execution The two execution skills shared ~70% of their content (task classification, fix loop severity, escalation rules, phase review logic, the final PR flow). Merge into a single `plan-execution` skill that auto-selects mode: - **Team mode** (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS set): worktrees, spawned teammates, shared task list, mailbox communication - **Subagent mode** (default): parallel dispatch via `run_in_background`, each subagent gets only what you pass it (lightweight context), independent `model` param per task Key corrections from the previous separate skills: - Subagents CAN run in parallel via `run_in_background` (not sequential) - Subagents are more efficient for well-scoped tasks (no context inheritance) - Team mode teammates inherit lead's model (noted as tradeoff) Also updates all references across CLAUDE.md, README.md, writing-plans, using-git-worktrees, systematic-debugging, implementer agent, and workflow-preferences hook. Removes /pr-create and /pr-merge references from README that were missed in previous commit. 716 lines (two files) → 502 lines (one file), -315 net across repo. https://claude.ai/code/session_01Ch89tRMukrCMshPhYskRnC
|
Claude finished @tombakerjr's task in 2m 3s —— View job Code Review CompleteLGTM - no blocking issues Reviewed the PR consolidation and pr-status rewrite. The changes improve maintainability while preserving functionality. All shell scripts are cross-platform compatible and defensively coded. Summary of Changes
CONSIDER (Non-blocking design tradeoffs)
Code Quality Checks
|
pr-create: Add existing-PR check using
gh pr list(exits 0 when no PRexists) instead of
gh pr view(exits non-zero). Make confirmation stepresilient with the same approach.
pr-status: Rewrite from snapshot-only to full readiness checker — watches
CI to completion, polls for review bot comment with timestamp validation
against the current CI run, scans for blocker keywords, and reports a
clear READY/NOT READY verdict.
pr-merge: Remove duplicated CI watching, comment polling, and blocker
scanning (Steps 3-6). Delegate readiness checking to /pr-status, keeping
only merge-specific logic: local verification, review feedback assessment,
fix loop, merge execution, and return to main.
Fixes #12
https://claude.ai/code/session_01Ch89tRMukrCMshPhYskRnC