skill_name: agent-pr-reviewer activation_code: AGENT_PR_REVIEWER_V1 version: 1.0.0 phase: any prerequisites:
- PR URL or PR number
- GitHub access (gh CLI authenticated) outputs:
- PR analysis report
- Agent-generated recommendation
- Structured summary for human review description: | Evaluates PRs containing agent changes using specialized agents to analyze quality, compliance, and improvement value. Provides turbobeest with a structured recommendation to streamline human review.
Agent PR Reviewer Skill
Purpose
When contributors submit PRs with agent improvements to turbobeest/dev-system, this skill:
- Extracts the agent changes from the PR
- Validates against
-01-agent-formatting/standards - Analyzes using specialized agents
- Summarizes changes and impact
- Recommends approve, request changes, or decline
Activation
/agent-pr-reviewer pr=123
Or with full URL:
/agent-pr-reviewer pr=https://github.com/turbobeest/dev-system/pull/123
Workflow
┌─────────────────────────────────────────────────────────────────────────┐
│ PR REVIEW WORKFLOW │
├─────────────────────────────────────────────────────────────────────────┤
│ │
│ Step 1: FETCH PR │
│ ├─ gh pr view {number} --json │
│ ├─ Extract changed files │
│ └─ Identify agent files (agents/**/*.md) │
│ ▼ │
│ Step 2: VALIDATE STRUCTURE │
│ ├─ Run agent-creation-validator on each changed agent │
│ ├─ Check tier compliance │
│ └─ Verify required sections present │
│ ▼ │
│ Step 3: ANALYZE CHANGES │
│ ├─ Diff analysis: what changed? │
│ ├─ Knowledge sources: added/removed/modified? │
│ ├─ Instructions: improved or degraded? │
│ └─ Potential regressions flagged │
│ ▼ │
│ Step 4: AGENT EVALUATION │
│ ├─ first-principles-engineer: Is this a good change? │
│ ├─ Domain expert (if applicable): Domain accuracy? │
│ └─ prd-auditor: Quality assessment │
│ ▼ │
│ Step 5: GENERATE RECOMMENDATION │
│ ├─ Synthesize agent opinions │
│ ├─ Flag concerns or conflicts │
│ └─ Provide structured recommendation │
│ │
└─────────────────────────────────────────────────────────────────────────┘
Output Format
╔═══════════════════════════════════════════════════════════════════════╗
║ PR REVIEW: #{number} ║
║ {pr-title} ║
╠═══════════════════════════════════════════════════════════════════════╣
║ ║
║ Author: {author} ║
║ Branch: {head} → {base} ║
║ Files Changed: {count} ({agent-count} agent files) ║
║ ║
╠═══════════════════════════════════════════════════════════════════════╣
║ AGENT CHANGES ║
╠═══════════════════════════════════════════════════════════════════════╣
║ ║
║ ┌────────────────────────────────────────────────────────────────┐ ║
║ │ MODIFIED: typescript-pro.md │ ║
║ ├────────────────────────────────────────────────────────────────┤ ║
║ │ + Added 3 knowledge sources │ ║
║ │ ~ Modified interpretive lens │ ║
║ │ ~ Updated 4 instructions │ ║
║ │ - Removed 1 deprecated reference │ ║
║ │ │ ║
║ │ Structure: ✓ Valid │ ║
║ │ Tier: Expert (unchanged) │ ║
║ │ Token Δ: +120 tokens (within budget) │ ║
║ └────────────────────────────────────────────────────────────────┘ ║
║ ║
║ ┌────────────────────────────────────────────────────────────────┐ ║
║ │ NEW: deno-expert.md │ ║
║ ├────────────────────────────────────────────────────────────────┤ ║
║ │ Category: backend-ecosystems/javascript-runtimes │ ║
║ │ Tier: Expert │ ║
║ │ Model: sonnet │ ║
║ │ │ ║
║ │ Structure: ✓ Valid │ ║
║ │ Curation Record: ✓ Present │ ║
║ │ Overlap Check: Low overlap with nodejs-expert │ ║
║ └────────────────────────────────────────────────────────────────┘ ║
║ ║
╠═══════════════════════════════════════════════════════════════════════╣
║ AGENT ANALYSIS ║
╠═══════════════════════════════════════════════════════════════════════╣
║ ║
║ first-principles-engineer: ║
║ "The typescript-pro changes add authoritative sources from ║
║ TypeScript's official handbook. The instruction modifications ║
║ improve specificity without losing generality. Recommend merge." ║
║ Confidence: HIGH ║
║ ║
║ typescript-pro (self-review of diff): ║
║ "New knowledge sources are accurate and high-authority. ║
║ Instruction changes align with TypeScript 5.x patterns. ║
║ No concerns." ║
║ Confidence: HIGH ║
║ ║
║ prd-auditor: ║
║ "Deno-expert is well-structured but missing explicit ║
║ differentiation from nodejs-expert in Identity section. ║
║ Minor improvement suggested." ║
║ Confidence: MEDIUM ║
║ ║
╠═══════════════════════════════════════════════════════════════════════╣
║ CONCERNS ║
╠═══════════════════════════════════════════════════════════════════════╣
║ ║
║ [LOW] deno-expert could clarify differentiation from nodejs-expert ║
║ ║
╠═══════════════════════════════════════════════════════════════════════╣
║ RECOMMENDATION ║
╠═══════════════════════════════════════════════════════════════════════╣
║ ║
║ ┌────────────────────────────────────────────────────────────────┐ ║
║ │ ✅ RECOMMEND APPROVE │ ║
║ │ │ ║
║ │ • typescript-pro improvements are high-quality │ ║
║ │ • deno-expert adds value with minor suggested tweak │ ║
║ │ • All structure validations pass │ ║
║ │ • Curation records present │ ║
║ └────────────────────────────────────────────────────────────────┘ ║
║ ║
║ Optional comment to author: ║
║ "Consider adding explicit differentiation from nodejs-expert in ║
║ deno-expert's Identity section (e.g., 'Deno-first approach with ║
║ native TypeScript, secure-by-default permissions model')." ║
║ ║
╚═══════════════════════════════════════════════════════════════════════╝
Recommendation Categories
| Recommendation | Meaning | Action |
|---|---|---|
| APPROVE | High-quality changes, ready to merge | turbobeest merges |
| APPROVE WITH SUGGESTIONS | Good changes, minor improvements optional | Merge now or wait for tweaks |
| REQUEST CHANGES | Issues that should be addressed | Author revises, re-review |
| DECLINE | Doesn't meet standards or conflicts with philosophy | Close with explanation |
Validation Checks
Structure Validation
- Frontmatter complete (name, tier, model, tools, modes)
- Identity section with interpretive lens
- Instructions section with Always/Mode-specific/Never
- Specializations defined
- Knowledge sources documented
- Output format specified
Quality Validation
- Tier classification appropriate
- Model selection justified
- Knowledge sources authoritative
- Instructions non-conflicting
- Vocabulary calibrated (15-20 terms)
Contribution Validation
- Curation record present (for new agents)
- No duplicate agents without differentiation
- Follows
-01-agent-formatting/standards - PR description explains rationale
Agent Panel
The review uses a panel of agents:
| Agent | Role | Focus |
|---|---|---|
first-principles-engineer | Quality arbiter | Is this fundamentally a good change? |
{domain-expert} | Domain accuracy | Are domain-specific details correct? |
prd-auditor | Standards compliance | Does it meet quality standards? |
architect-reviewer | Structural coherence | Does the agent fit the ecosystem? |
Integration
GitHub CLI Commands Used
# Fetch PR details
gh pr view {number} --json title,author,body,files,headRefName,baseRefName
# Get diff
gh pr diff {number}
# Add review comment (after human approval)
gh pr review {number} --approve --body "..."
gh pr review {number} --request-changes --body "..."
Signals
| Signal | Meaning |
|---|---|
PR_REVIEW_STARTED | Analysis begun |
PR_VALIDATION_COMPLETE | Structure checks done |
PR_ANALYSIS_COMPLETE | Agent analysis done |
PR_RECOMMENDATION_READY | Ready for human review |
Human Override
The skill provides a recommendation, but turbobeest makes the final decision.
After reviewing the analysis:
[A]Accept recommendation and merge[R]Request changes (with specific feedback)[D]Decline PR[O]Override recommendation with rationale
All decisions are logged in .claude/pr-reviews/ for accountability.