pr-review - SKILL.md Agent Skill

name: pr-review description: Review recent PRs with deep context and label-specific checks argument-hint: '[count=10]' allowed-tools: Bash, Read, Grep, Glob, Skill

PR Review

Fetch recent PRs, classify, read source code, and produce contextual reviews.

Input

$ARGUMENTS — Number of PRs (default: 10)

Examples:

/pr-review — 10 most recent open PRs
/pr-review 20 — 20 PRs

Workflow

1. Fetch recent PRs

gh pr list --repo toss/es-toolkit --state open --limit {count} --json number,title,author,labels,createdAt

2. Triage each PR

Run /pr-triage {number} for each PR to classify.

3. Deep review per PR

For each PR, go beyond classification — read the actual source code and provide context:

a. Read current code

Read the files being modified on main to understand the current behavior:

Function source: what it does, how it works
Existing tests: what's already covered
JSDoc: documented behavior and examples

b. Understand the PR change

Read the diff to understand what's changing:

What's the before vs after?
Why does the contributor say this change is needed?
Does the change match what the code actually needs?

c. Assess impact

Breaking changes: Does the function signature change? Will existing users be affected?
JS spec alignment: Does current behavior follow JS/lodash conventions? Does the PR break that?
Scope: Are all affected files updated? (e.g., docs in all 4 languages, related functions)
Design principles: Does this align with es-toolkit's 85% use case / simplicity philosophy?

d. Label-specific checks

compat-fix: Run /compat-review {number}
new-function: Check completeness (impl + spec + re-export + 4 lang docs)
core-change: yarn vitest run src/{category}/{fn}.spec.ts + review breaking impact
docs: Check all 4 language files updated consistently
types: tsc --noEmit

4. Detect duplicates

Group PRs by same function name or same files modified.

5. Report per PR

For each PR, output:

### PR #{number} — {title}

**Classification**: {labels}
**Changed files**: {summary}

**Current code**: {what the function does now, with code snippet}
**PR change**: {what's changing and why, with before/after}
**Context**: {JS spec, lodash behavior, design principle considerations}

**Checkpoints**:
- {specific things to verify}

**Verdict**: {merge / merge with changes / hold for discussion / reject}

6. Summary

## PR Review — {date}

| # | Title | Label | Verdict |
|---|-------|-------|---------|

- {N} ready to merge
- {N} need changes
- {N} need discussion
- {N} potential duplicates