Pr Review logo

Pr Review

Organization
agentic-community
pr-review

Review a GitHub pull request using multiple expert personas. Takes a PR URL as input, analyzes the changes, and generates comprehensive review feedback from different perspectives (Merge Specialist, Frontend, Backend, Security, DevOps, AI/Agent, SRE, Chief Architect).

Overview

Publisheragentic-community
Repositorymcp-gateway-registry
Skill namepr-review
Stars
969
Forks
244
Bundled files
9
LicenseApache-2.0
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 9 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by agentic-community on GitHub. Read the source before you install it.

Installation

Install the Pr Review AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/agentic-community/mcp-gateway-registry/tree/main/.claude/skills/pr-review
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/agentic-community/mcp-gateway-registry.git /tmp/mcp-gateway-registry
mkdir -p .claude/skills
cp -r /tmp/mcp-gateway-registry/.claude/skills/pr-review .claude/skills/pr-review
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Pr Review in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Pr Review on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Pr Review is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

PR Review Skill

Use this skill to review GitHub pull requests comprehensively using multiple expert personas. Each persona brings specialized knowledge to identify issues from different perspectives.

Input

The skill takes a GitHub PR URL as input:

  • Format: https://github.com/{owner}/{repo}/pull/{number}
  • Example: https://github.com/agentic-community/mcp-gateway-registry/pull/123

Output

Creates review documentation in .scratchpad/pr-{pr-number}/ containing:

  • review.md - Comprehensive review from all personas

Workflow

Step 1: Parse PR URL and Fetch PR Details

  1. Extract the PR number from the URL
  2. Use gh pr view {number} to get PR details
  3. Use gh pr diff {number} to get the changes
  4. Identify which files are changed and their types (frontend, backend, etc.)

Step 2: Determine Relevant Personas

Based on the files changed, determine which personas should review:

Changed FilesPersonas to Engage
/frontend/**Merge Specialist, Frontend Developer, Chief Architect
/registry/**Merge Specialist, Backend Developer, Security Engineer, SRE, Chief Architect
/registry/core/config.py, /registry/api/config_routes.pyMerge Specialist, Backend Developer, DevOps Engineer, Security Engineer, Chief Architect
/auth_server/**Merge Specialist, Backend Developer, Security Engineer, Chief Architect
/terraform/**, /charts/**, /docker/**Merge Specialist, DevOps Engineer, SRE, Chief Architect
/agents/**, /servers/**Merge Specialist, AI/Agent Developer, Backend Developer, Chief Architect
/metrics-service/**Merge Specialist, SRE Engineer, Backend Developer, Chief Architect
*.md, docs/**Merge Specialist, Chief Architect
pyproject.toml, requirements*.txtMerge Specialist, DevOps Engineer, Security Engineer, Chief Architect
tests/**Merge Specialist, Backend Developer, Chief Architect
.env.exampleMerge Specialist, DevOps Engineer, Chief Architect

Note: Merge Specialist and Chief Architect always participate in every review.

Step 2.5: Detect New or Modified Configuration Parameters (CRITICAL)

Before running any reviews, determine whether this PR introduces or modifies any configuration parameters across the three deployment surfaces. If it does, the unified parameter reference must be updated in the same PR. A missing update is a blocker.

Detection command:

bash
# Any diff that touches one of the canonical parameter-carrying files triggers this check
gh pr diff {pr-number} --name-only | grep -E \
  -e '^\.env\.example$' \
  -e '^docker-compose(\.|$)' \
  -e '^terraform/aws-ecs/terraform\.tfvars\.example$' \
  -e '^terraform/aws-ecs/variables\.tf$' \
  -e '^terraform/aws-ecs/modules/.+/(variables|ecs-services)\.tf$' \
  -e '^charts/.+/values\.yaml$' \
  -e '^charts/.+/templates/(deployment|secret)\.yaml$' \
  -e '^registry/core/config\.py$' \
  -e '^registry/api/config_routes\.py$'

If any files match, every new or renamed parameter must be reflected in docs/unified-parameter-reference.md. Verify with:

bash
# For each new parameter name, confirm the reference file mentions it
for PARAM in $(gh pr diff {pr-number} | grep -E '^\+[A-Z_]{3,}=' | sed 's/^+//;s/=.*//' | sort -u); do
  if ! grep -q "$PARAM" docs/unified-parameter-reference.md; then
    echo "MISSING from unified-parameter-reference.md: $PARAM"
  fi
done

Merge-blocking checks (add to the Merge Specialist review section):

  • docs/unified-parameter-reference.md is included in the diff whenever any parameter-carrying file is touched.
  • Every new .env variable has a row, with Docker / Terraform / Helm columns filled (or explicitly blank with a justification in the PR description).
  • Every new Terraform variable appears in the reference.
  • Every new Helm value appears in the reference.
  • Secrets are flagged with (secret).
  • New rows live in an existing logical group, or the PR adds a new group with a clear rationale.
  • Renamed parameters have the old row updated in place (not a duplicate).
  • Deleted parameters have their row removed (not left stale).
  • registry/api/config_routes.py CONFIG_GROUPS is updated so the parameter surfaces in GET /api/config/full and the System Config UI.

If any of the above is missing, the review verdict is REQUEST CHANGES with a blocker titled "Unified parameter reference not updated".

Step 2.6: Detect New or Changed API Endpoints (CRITICAL)

api/openapi.json is the published API contract, and it is hand-refreshed: no script generates it and no CI job checks it. A PR that adds a route therefore passes every gate while leaving the new endpoint invisible to API consumers. This has already happened, more than once in a row, so the spec shipped missing endpoints from several merged PRs at the same time.

Detection: does the diff add or change a route decorator?

bash
# Matches ANY router object, not just `router`/`app`: real routes are declared on
# names like `cimd_router`, `wellknown_router`, and a narrower pattern misses them
# (verified: it missed #1711's `@cimd_router.get(...)` entirely).
gh pr diff {pr-number} | grep -E '^\+\s*@[A-Za-z_][A-Za-z0-9_]*\.(get|post|put|patch|delete)\('

If that matches, check whether the spec was refreshed:

bash
gh pr diff {pr-number} --name-only | grep -q '^api/openapi\.json$' \
  && echo "spec refreshed" || echo "SPEC NOT REFRESHED"

And confirm the new path actually landed in it, rather than the file being touched for an unrelated reason:

bash
git show {pr-branch}:api/openapi.json | python3 -c "
import json, sys
print(sorted(json.load(sys.stdin)['paths']))
" | tr ',' '\n' | grep -i '<the new path>'

Merge-blocking checks:

  • Every route the diff adds appears in api/openapi.json.
  • Every route the diff removes is gone from it.
  • info.version is a clean release semver, not the app's git-describe development string (1.30.0-46-g...).
  • The diff to api/openapi.json is confined to the affected paths. A wholesale reformat usually means it was regenerated with ensure_ascii=False, which rewrites every non-ASCII character and hides the real change.
  • Nothing was removed unexpectedly. Unexplained removals usually mean the container was built with a feature flag off or the wrong DEPLOYMENT_MODE, so a whole router never registered.

A route added behind a default-off feature flag still belongs in the spec: FastAPI registers the route regardless, and the flag only changes the response at request time.

If a route was added and the spec was not refreshed, the verdict is REQUEST CHANGES with a blocker titled "OpenAPI spec not refreshed for new endpoint". The procedure is in AGENTS.md; refreshing it can also be offered as a follow-up PR when the author would rather not rebuild locally.

Step 3: Run Tests and Quality Checks

Before reviewing, run the test suite to verify the PR doesn't break anything:

bash
# Checkout the PR
gh pr checkout {pr-number}

# Run tests
uv run pytest tests/ -n 8 --tb=short

# Run linting
uv run ruff check . && uv run ruff format --check .

# Run security scan (if applicable)
uv run bandit -r registry/ auth_server/ -q

# Return to main branch when done
git checkout main

Step 4: Create Review Folder

Create the folder structure:

.scratchpad/pr-{pr-number}/
└── review.md

Step 5: Conduct Multi-Persona Review

For each relevant persona, adopt that perspective and review the changes. Reference the persona definition files:

Theory check (always): the Chief Architect persona must read Theory of the System and walk the diff against its "how to change this system without breaking its theory" checklist. If the PR violates a core invariant (control-plane/data-plane split, generic gateway, A2A peer-to-peer, mode axes, config parity, fail-closed admission, IdP-agnosticism, MCP spec compliance) without explicitly arguing for the change, flag it to the user as a blocker.

Step 6: Write Comprehensive Review (review.md)

Generate the review document using this structure:

markdown
# PR Review: #{pr-number} - {pr-title}

*Review Date: {date}*
*PR URL: {pr-url}*
*Author: {author}*

## PR Summary

{Brief description of what the PR does based on PR description and changes}

### Files Changed

| File | Type | Lines Added | Lines Removed |
|------|------|-------------|---------------|
| {file} | {type} | +{n} | -{n} |

### Test Results

| Check | Status | Details |
|-------|--------|---------|
| Unit Tests | {PASS/FAIL} | {summary} |
| Integration Tests | {PASS/FAIL} | {summary} |
| Linting | {PASS/FAIL} | {summary} |
| Security Scan | {PASS/FAIL} | {summary} |

### API Spec Check

*Only required when the diff adds or changes a route decorator (see Step 2.6). Mark "Not Applicable" otherwise.*

| Check | Status | Details |
|-------|--------|---------|
| Every added route appears in `api/openapi.json` | {PASS/FAIL/N/A} | {the paths} |
| Every removed route is gone from it | {PASS/FAIL/N/A} | n/a |
| `info.version` is a release semver, not a dev string | {PASS/FAIL/N/A} | n/a |
| Spec diff confined to the affected paths | {PASS/FAIL/N/A} | n/a |

### Configuration Parameter Surface Check

*Only required when the PR touches any parameter-carrying file (see Step 2.5). Mark "Not Applicable" if the detection command returned no matches.*

| Check | Status | Details |
|-------|--------|---------|
| Unified parameter reference updated (`docs/unified-parameter-reference.md`) | {PASS/FAIL/N/A} | {list of new/renamed/removed parameter names and which rows were added} |
| Docker column populated (`.env.example`, `docker-compose*.yml`) | {PASS/FAIL/N/A} |, |
| Terraform column populated (`variables.tf`, `terraform.tfvars.example`, module wiring) | {PASS/FAIL/N/A} |, |
| Helm column populated (`charts/.../values.yaml`, stack values, templates) | {PASS/FAIL/N/A} |, |
| `registry/api/config_routes.py` `CONFIG_GROUPS` updated | {PASS/FAIL/N/A} |, |
| Secrets flagged with **(secret)** and wired through Secrets Manager / `secretKeyRef` | {PASS/FAIL/N/A} |, |

---

## Review Panel

| Role | Reviewer | Verdict |
|------|----------|---------|
| Merge Specialist | Gatekeeper | {verdict} |
| {Role} | {Name} | {verdict} |
| Chief Architect | Atlas | {verdict} |

---

{Include each relevant persona's review section using the format from their persona file}

---

## Review Summary

| Reviewer | Verdict | Blockers | Key Concerns |
|----------|---------|----------|--------------|
| {Reviewer} | {verdict} | {count} | {summary} |

### Blockers (Must Fix)

1. {Blocker description}
   - Raised by: {persona}
   - File: `{file:line}`
   - Fix: {suggested fix}

### Should Fix (Important)

1. {Issue description}
   - Raised by: {persona}
   - File: `{file:line}`
   - Recommendation: {suggestion}

### Consider (Nice to Have)

1. {Suggestion}
   - Raised by: {persona}

---

## Final Recommendation

**Overall Verdict: {APPROVE / APPROVE WITH CHANGES / REQUEST CHANGES}**

### Required Actions Before Merge

- [ ] {Action 1}
- [ ] {Action 2}
- [ ] (If config params changed) `docs/unified-parameter-reference.md` updated and all three surface columns consistent with the diff

### Post-Merge Actions

- [ ] {Action 1}

Step 7: Render the review as HTML

A review document is long and heavily tabular, which reads badly as raw markdown in a terminal. Render it so the reader can open a formatted page instead:

bash
uv run python scripts/render-doc-html.py .scratchpad/pr-NNNN/review.md

That writes review.html beside the markdown, self-contained (all CSS inline, no CDN, no JavaScript, renders from file://), using the same stylesheet as the explainer skill so every generated document looks the same. Title and byline come from the document's H1 and the italic lines under it, and a section nav is built from the H2 headings.

Useful flags:

  • --footer-html '...' for provenance, such as the commit the review was performed against.
  • --diagrams DIR to inline SVG from DIR/<key>.svg wherever the markdown fences a block as ```svg:<key> Optional caption. Keep the ASCII inside the fence: it is what a terminal reader sees, and a missing .svg file falls back to it rather than losing the diagram. A review rarely needs this; a design document often does.
  • --code-style invert for the template's dark code blocks. The default (match) makes code blocks follow the page surface, which suits a document that is mostly code.

Then check the output, because an unparsed HTML file is worse than none:

bash
uv run python scripts/prose-scan.py --strict .scratchpad/pr-NNNN/review.md .scratchpad/pr-NNNN/review.html

The renderer already warns about unfilled placeholders, broken in-page anchors, and svg: fences with no matching file. Fix anything it reports and re-run.

Re-render after every edit to the markdown. The markdown is the source; the HTML is a build artifact, and the two drift the moment you hand-edit the HTML.

Open it for the reader rather than starting a server:

bash
code -r .scratchpad/pr-NNNN/review.html

Do not start a server. The HTML is self-contained, so the editor's preview or a downloaded copy is enough, and a process the user did not ask for is one they have to hunt down later. If they want HTTP, offer this and let them run it in a VS Code integrated terminal, which is what makes VS Code forward the port:

bash
python3 -m http.server 8112 --bind 127.0.0.1 --directory /abs/path/to/.scratchpad/pr-NNNN

Then the URL is http://127.0.0.1:8112/review.html, or drop the filename for a directory listing. Point --directory at the single document's folder, never at .scratchpad/ itself: that folder holds credential files and http.server serves everything below its root.

Trust model for the generated HTML

The renderer treats the markdown body, the byline derived from it, and any inlined SVG as untrusted, because this skill summarizes GitHub-fetched content into that markdown. Raw HTML in the markdown is disabled, the byline is escaped with an href scheme allowlist, and an SVG carrying a script, an event handler, or an external reference aborts the render. A <script> in a PR body therefore renders as visible, inert text.

--byline-html and --footer-html are the two exceptions: both are inserted verbatim. Use them only for first-party provenance text you wrote. Never pass a PR title, an issue body, an author name, or any other fetched metadata through either one.

Step 8: Present Review Summary

After creating the review document, present a summary to the user:

  1. Display the overall verdict
  2. List any blockers that must be addressed
  3. Provide the path to the full review document
  4. Offer to explain any specific findings in detail
  5. End the response with a clickable link to the PR (see below)
Always end with a clickable PR link

The last line of every review response must be a full markdown link to the PR, so the reader can open it without copying a number into a URL bar:

markdown
[#1711, feat: CIMD client metadata document](https://github.com/agentic-community/mcp-gateway-registry/pull/1711)

Rules:

  • Use the full https://github.com/{owner}/{repo}/pull/{number} URL. A bare #1711 renders as plain text in the terminal and is not clickable.
  • Include the PR title after the number, so the link says what it points at.
  • This applies to every response about the PR, not just the first one: follow-up answers, "pushed the fixes", and status updates all end with the link.
  • If the response covers more than one PR, link each of them.

Review Principles

From CLAUDE.md

  • Simplicity: Code should be maintainable by entry-level developers
  • No Over-engineering: Only make changes that are directly requested
  • Security First: Check for OWASP vulnerabilities, proper input validation
  • Test Coverage: Verify tests exist for new functionality
  • Documentation: Ensure docstrings and comments are appropriate

Severity Levels

  • Blocker: Must be fixed before merge (security vulnerabilities, failing tests, breaking changes)
  • Major: Should be fixed before merge (code quality issues, missing tests)
  • Minor: Nice to fix (style issues, documentation improvements)

Verdict Criteria

APPROVE:

  • All tests pass
  • No security vulnerabilities
  • Code quality meets standards
  • No breaking changes (or justified)

APPROVE WITH CHANGES:

  • Minor issues that should be addressed
  • No blockers
  • Author can address and merge

REQUEST CHANGES:

  • Failing tests
  • Security vulnerabilities
  • Breaking changes without justification
  • Significant code quality issues

Example Usage

User: "/pr-review https://github.com/agentic-community/mcp-gateway-registry/pull/456"

  1. Parse URL: PR #456
  2. Fetch PR details and diff
  3. Identify changed files: registry/routes/auth.py, tests/unit/test_auth.py
  4. Determine personas: Merge Specialist, Backend Developer, Security Engineer, Chief Architect
  5. Run tests: All pass
  6. Create .scratchpad/pr-456/review.md
  7. Conduct reviews from each persona
  8. Render review.html with scripts/render-doc-html.py
  9. Present summary with verdict

Notes

  • Always run tests before reviewing to ensure baseline quality
  • Focus review effort on areas most relevant to the changed files
  • Be constructive and specific - provide file/line references
  • Acknowledge good practices, not just problems
  • Consider the author's experience level when phrasing feedback
  • The markdown is the source of truth and the HTML is a build artifact. Re-render after any edit rather than hand-editing review.html

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Pr Review AI skill do?

Review a GitHub pull request using multiple expert personas. Takes a PR URL as input, analyzes the changes, and generates comprehensive review feedback from different perspectives (Merge Specialist, Frontend, Backend, Security, DevOps, AI/Agent, SRE, Chief Architect).

Why use Pr Review on TypingMind?

Because you install it once and use it with any model. Pr Review is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Pr Review in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/agentic-community/mcp-gateway-registry/tree/main/.claude/skills/pr-review. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Pr Review?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Pr Review?

As many as you like. As long as a model supports skills, you can use Pr Review with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Pr Review AI skill free?

Yes. It is published on GitHub by agentic-community under the Apache-2.0 license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇