personal-parallel-code-review

name	personal-parallel-code-review
description	Review multiple Flutter repository pull requests in parallel using subagents, each in its own isolated shared workspace/worktree.

Personal Parallel Code Review

This skill guides you through reviewing multiple Flutter pull requests concurrently by delegating individual PR reviews to subagents. Each subagent runs in its own isolated shared workspace (comparable to a git worktree), preventing conflict between concurrent checkouts and test executions.

Prerequisites

• GitHub CLI (gh) installed and authenticated.
• Local clone of the flutter/flutter repository.
• Subagents enabled with access to write/execute tools.

Workflow

1. Cleanup Stale Workspaces & Artifacts

Before starting a new parallel review session, clean up any leftover processes, workspaces, and artifacts from previous sessions:

1. Kill running subagents: Use the manage_subagents tool with Action: 'kill_all' to terminate any active subagents from prior runs (this automatically deletes their corresponding shared workspaces).
2. Prune git worktrees: Run git worktree prune in the git repository to clean up any stale worktree metadata.

   [!NOTE] If the sandbox blocks git commands (like checkouts or pruning) due to stale worktree paths pointing to the deleted App Data directory, manually delete the stale metadata directories from .git/worktrees/ (e.g. rm -rf .git/worktrees/subagent-*).

3. Delete old review branches: Run git branch -D on any local branches matching review-* if they are no longer needed.
4. Archiving previous artifacts: Remove or rename previous review report files (like comprehensive_parallel_code_review_report.md or individual subagent report files) in the parent workspace to ensure clean results.

2. Identify target PRs

Choose the list of Pull Request numbers to review (typically 3 to 5 PRs). You can retrieve open PRs using:

gh pr list --limit 5 --state open

3. Spawn Parallel Subagents

For each PR, invoke a subagent of type self with the Workspace parameter set to share. This assigns each subagent an isolated git worktree-like workspace.

Specify the following instructions in each subagent prompt:

• Use the personal-flutter-code-review skill.
• Checkout the specific PR: gh pr checkout <PR_NUMBER> -b review-<PR_NUMBER>.
• Inspect the PR changes and description using gh pr view and gh pr diff.
• Apply the Flutter Code Review Checklist.
• Check formatting and run relevant tests or write regression/verification tests if appropriate.

  [!TIP] Since standard sandboxed execution blocks outbound network access and reads to global caches outside the workspace (like /Users/katelovett/.pub-cache), run tests either:

  1. By setting BypassSandbox: true (which prompts the user to approve).
  2. By configuring the PUB_CACHE environment variable to a folder inside the workspace and running flutter test --offline.

• Document the findings, categorizing by Correctness, Breaking Changes, Performance, Documentation, and Testing.
• Report the results back to the orchestrating parent agent.

3. Wait and Orchestrate

As subagents perform the reviews, the parent agent remains in standby. Once the subagents report back, gather their individual summaries.

4. Generate Comprehensive Report

Assemble the results from all parallel reviews into a structured markdown report that includes:

• Executive Summary: Overall state of the reviewed PRs (e.g., how many are ready to merge, how many have issues, etc.).
• Detailed PR Reviews: A dedicated section for each PR:
  • PR details (title, number, author, branch).
  • Findings summary (issues, styling, performance).
  • Test verification results.
  • Actionable recommendation (Approve, Request Changes, or Needs More Info).
• Common Trends: Any common antipatterns, shared code areas affected, or process recommendations.