Skip to content
Agent Memory
Command

/pr-reviews

Review open pull requests for code quality, project alignment, and risks.

From plugin
mulch
3224 skills4 commands

How it fires

How this command gets triggered: by you, by Claude, or both.

  • Fires itselfClaude auto-loads it when your prompt matches the work.
  • You can call itInvoke it directly when you want it.
  • Slash command/pr-reviews

Context preview

What this command does when you run it.

Review open pull requests for code quality, project alignment, and risks.

Command definition

pr-reviews.md
name: pr-reviews

intro

Review open pull requests for code quality, project alignment, and risks.

**Argument:** `$ARGUMENTS` — optional PR number(s) to review (e.g., `9` or `9 12 15`). If empty, review all open PRs.

Steps

1. Discover PRs to review

  • If `$ARGUMENTS` contains PR number(s), use those
  • Otherwise, run `gh pr list --state open --json number,title,author,headRefName,additions,deletions` to get all open PRs
  • If there are no open PRs, say so and stop

2. Spawn a review team

Use the Task tool to spawn parallel agents (one per PR). Each agent should:

a. Gather context

  • `gh pr view <number> --json title,body,author,additions,deletions,files,commits,comments,reviews,headRefName,baseRefName`
  • `gh pr diff <number>` to get the full diff
  • Read any files touched by the PR to understand the surrounding code

b. Code quality review

  • Check for correctness — does the code do what the PR claims?
  • Check for bugs, edge cases, and error handling gaps
  • Check adherence to project conventions (see CLAUDE.md): strict TypeScript, zero runtime deps, Biome formatting, tab indentation, 100-char line width
  • Check test coverage — are new code paths tested? Do tests follow the "never mock what you can use for real" philosophy?
  • Flag any security concerns (injection, unsafe input handling, etc.)

c. Project alignment review

  • Does this change fit the project's architecture and direction?
  • Does it follow existing patterns or introduce unnecessary new ones?
  • Is the scope appropriate — does it do too much or too little?
  • Are there breaking changes or backward-compatibility concerns?

d. Risk assessment

  • What could go wrong if this is merged?
  • Are there performance implications?
  • Does it touch critical paths (config loading, expertise storage, JSONL integrity)?
  • Are there dependency or compatibility risks?
  • Could it conflict with other open PRs?

e. Produce a review summary

Each agent should return a structured review:

  • **PR:** `#<number> — <title>` by `<author>`
  • **Verdict:** Approve / Request Changes / Needs Discussion
  • **Summary:** 2-3 sentence overview
  • **Strengths:** What's good about this PR
  • **Issues:** Bugs, risks, or concerns (with file:line references)
  • **Suggestions:** Non-blocking improvements
  • **Project alignment:** How well it fits mulch's direction

3. Present consolidated report

After all agents complete, present a single consolidated report with:

  • A summary table of all reviewed PRs with verdicts
  • The detailed review for each PR
  • Any cross-PR concerns (conflicts, overlapping changes, pattern inconsistencies)
  • Recommended merge order if multiple PRs are ready
Read more
Ships withmulch

Structured expertise management for AI agent workflows. Agents start every session from zero. The pattern your agent discovered yesterday is forgotten today. Mulch fixes this: agents call ml record to write learnings, and ml query to read them.

Get the whole plugin
Stats
328
Stars
41
Forks
Active
Maintenance
TypeScript
Language
MIT
License
6d ago
Last commit
6mo ago
Created

Repo: jayminwest/mulch