Add Stage 2 secure review workflow for external PRs

- Runs after Stage 1 via workflow_run trigger
- Has access to repository secrets
- Downloads PR artifact and performs review
- Maintains security by never checking out fork code
This commit is contained in:
Rasmus Widing 2025-08-19 10:09:45 +03:00
parent d64745991b
commit fc97b4f5bd

385
.github/workflows/claude-review-ext.yml vendored Normal file
View File

@ -0,0 +1,385 @@
name: Claude Review External - Stage 2 (Secure Review)
# This workflow runs after Stage 1 completes
# It has access to secrets and performs the actual Claude review
# Security: Runs in the context of the base repository, not the fork
on:
workflow_run:
workflows: ["Claude Review External - Stage 1 (PR Info Collection)"]
types: [completed]
jobs:
check-authorization:
# Only run if Stage 1 completed successfully
if: github.event.workflow_run.conclusion == 'success'
runs-on: ubuntu-latest
outputs:
authorized: ${{ steps.check.outputs.authorized }}
pr_number: ${{ steps.check.outputs.pr_number }}
comment_author: ${{ steps.check.outputs.comment_author }}
trigger_phrase: ${{ steps.check.outputs.trigger_phrase }}
permissions:
actions: read
pull-requests: read
steps:
- name: Download PR Info Artifact
uses: actions/github-script@v7
id: download-artifact
with:
script: |
// Get artifacts from the workflow run
const artifacts = await github.rest.actions.listWorkflowRunArtifacts({
owner: context.repo.owner,
repo: context.repo.repo,
run_id: ${{ github.event.workflow_run.id }}
});
const matchArtifact = artifacts.data.artifacts.find(artifact =>
artifact.name.startsWith('pr-info-')
);
if (!matchArtifact) {
core.setFailed('No PR info artifact found');
return;
}
// Download the artifact
const download = await github.rest.actions.downloadArtifact({
owner: context.repo.owner,
repo: context.repo.repo,
artifact_id: matchArtifact.id,
archive_format: 'zip'
});
const fs = require('fs');
const path = require('path');
fs.writeFileSync('pr-info.zip', Buffer.from(download.data));
// Extract the artifact
const { exec } = require('child_process');
await new Promise((resolve, reject) => {
exec('unzip -o pr-info.zip', (error, stdout, stderr) => {
if (error) reject(error);
else resolve(stdout);
});
});
console.log('Artifact downloaded and extracted');
- name: Check Authorization
id: check
run: |
# Read PR info
PR_INFO=$(cat pr-info.json)
PR_NUMBER=$(echo "$PR_INFO" | jq -r '.prNumber')
COMMENT_AUTHOR=$(echo "$PR_INFO" | jq -r '.commentAuthor // .prAuthor')
TRIGGER_PHRASE=$(echo "$PR_INFO" | jq -r '.triggerPhrase // ""')
EVENT_NAME=$(echo "$PR_INFO" | jq -r '.eventName')
echo "PR Number: $PR_NUMBER"
echo "Comment Author: $COMMENT_AUTHOR"
echo "Trigger Phrase: $TRIGGER_PHRASE"
echo "Event: $EVENT_NAME"
# Check if this was triggered by a comment with the right phrase
AUTHORIZED="false"
if [[ "$EVENT_NAME" == "pull_request" ]]; then
# Automatic review for all PRs
AUTHORIZED="true"
echo "Authorized: Automatic review for PR event"
elif [[ "$TRIGGER_PHRASE" == "@claude-review-ext" ]]; then
# Check if user is authorized
AUTHORIZED_USERS='["Wirasm", "coleam00", "sean-eskerium"]'
if echo "$AUTHORIZED_USERS" | jq -e --arg user "$COMMENT_AUTHOR" 'contains([$user])' > /dev/null; then
AUTHORIZED="true"
echo "Authorized: $COMMENT_AUTHOR is in authorized users list"
else
echo "Not authorized: $COMMENT_AUTHOR is not in authorized users list"
fi
fi
echo "authorized=$AUTHORIZED" >> $GITHUB_OUTPUT
echo "pr_number=$PR_NUMBER" >> $GITHUB_OUTPUT
echo "comment_author=$COMMENT_AUTHOR" >> $GITHUB_OUTPUT
echo "trigger_phrase=$TRIGGER_PHRASE" >> $GITHUB_OUTPUT
claude-review:
needs: check-authorization
if: needs.check-authorization.outputs.authorized == 'true'
runs-on: ubuntu-latest
permissions:
contents: read # Read repository contents
pull-requests: write # Comment on PRs
issues: write # Comment on issues
actions: read # Read CI results
id-token: write # Required for OIDC authentication
steps:
- name: Checkout Repository (Base Branch)
uses: actions/checkout@v4
with:
# SECURITY: Checkout base branch, not PR code
ref: ${{ github.event.workflow_run.head_branch }}
fetch-depth: 0
- name: Download PR Info
uses: actions/github-script@v7
with:
script: |
// Get artifacts from the workflow run
const artifacts = await github.rest.actions.listWorkflowRunArtifacts({
owner: context.repo.owner,
repo: context.repo.repo,
run_id: ${{ github.event.workflow_run.id }}
});
const matchArtifact = artifacts.data.artifacts.find(artifact =>
artifact.name.startsWith('pr-info-')
);
if (!matchArtifact) {
core.setFailed('No PR info artifact found');
return;
}
// Download the artifact
const download = await github.rest.actions.downloadArtifact({
owner: context.repo.owner,
repo: context.repo.repo,
artifact_id: matchArtifact.id,
archive_format: 'zip'
});
const fs = require('fs');
fs.writeFileSync('pr-info.zip', Buffer.from(download.data));
// Extract the artifact
const { exec } = require('child_process');
await new Promise((resolve, reject) => {
exec('unzip -o pr-info.zip', (error, stdout, stderr) => {
if (error) reject(error);
else resolve(stdout);
});
});
// Read and parse PR info
const prInfo = JSON.parse(fs.readFileSync('pr-info.json', 'utf8'));
// Save key information as environment variables
core.exportVariable('PR_NUMBER', prInfo.prNumber);
core.exportVariable('PR_TITLE', prInfo.prTitle);
core.exportVariable('PR_AUTHOR', prInfo.prAuthor);
core.exportVariable('HEAD_SHA', prInfo.headSha);
console.log(`Loaded PR #${prInfo.prNumber} information`);
- name: Fetch PR Branch for Analysis
run: |
# Fetch the PR branch to analyze (but don't checkout)
git fetch origin pull/${{ env.PR_NUMBER }}/head:pr-branch
# Create a safe diff for analysis
git diff origin/${{ github.event.workflow_run.head_branch }}...pr-branch > pr-diff.patch
echo "Fetched PR branch for analysis (not checked out for security)"
- name: Post Starting Comment
uses: actions/github-script@v7
with:
script: |
await github.rest.issues.createComment({
owner: context.repo.owner,
repo: context.repo.repo,
issue_number: ${{ env.PR_NUMBER }},
body: `🤖 **Claude Review (External PR) Starting...**\n\nAnalyzing PR changes in a secure environment. This may take a few minutes.`
});
- name: Run Claude Code Review
id: claude
uses: anthropics/claude-code-action@beta
timeout-minutes: 15
with:
claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
# Use the external review trigger phrase
trigger_phrase: "@claude-review-ext"
# Custom context for PR review
pr_number: ${{ env.PR_NUMBER }}
# Review-specific instructions (same as claude-review.yml)
custom_instructions: |
You are performing a CODE REVIEW for an EXTERNAL PULL REQUEST.
This is a secure two-stage review process for PRs from forks.
## Security Context
You are running in a secure environment with access to repository secrets.
The PR code has been fetched but NOT checked out for security reasons.
Review the changes in pr-diff.patch file.
## Your Role
You are reviewing code for Archon V2 Alpha, a local-first AI knowledge management system in early alpha stage.
## Architecture Context
- Frontend: React + TypeScript + Vite (port 3737)
- Backend: FastAPI + Socket.IO + Python (port 8181)
- MCP Service: MCP protocol server (port 8051)
- Agents Service: PydanticAI agents (port 8052)
- Database: Supabase (PostgreSQL + pgvector)
## Review Process
1. **Understand Changes**
- Review the pr-diff.patch file to see all changes
- Check what files were changed and understand the context
- Analyze the impact across all services (frontend, backend, MCP, agents)
- Consider interactions between components
## Review Focus Areas
### 1. Code Quality - Backend (Python)
- Type hints on all functions and classes
- Pydantic v2 models for data validation (ConfigDict not class Config, model_dump() not dict())
- No print() statements (use logging instead)
- Proper error handling with detailed error messages
- Following PEP 8
- Google style docstrings where appropriate
### 2. Code Quality - Frontend (React/TypeScript)
- Proper TypeScript types (avoid 'any')
- React hooks used correctly
- Component composition and reusability
- Proper error boundaries
- Following existing component patterns
### 3. Structure & Architecture
- Each feature self-contained with its own models, service, and tools
- Shared components only for things used by multiple features
- Proper separation of concerns across services
- API endpoints follow RESTful conventions
### 4. Testing
- Unit tests co-located with code in tests/ folders
- Edge cases covered
- Mocking external dependencies
- Frontend: Vitest tests for components
- Backend: Pytest tests for services
### 5. Alpha Project Principles (from CLAUDE.md)
- No backwards compatibility needed - can break things
- Fail fast with detailed errors (not graceful failures)
- Remove dead code immediately
- Focus on functionality over production patterns
### 6. Security Considerations for External PRs
- Check for any attempts to access secrets or environment variables
- Look for suspicious network requests or data exfiltration
- Verify no hardcoded credentials or API keys
- Check for SQL injection vulnerabilities
- Review any file system operations
## Required Output Format
## Summary
[2-3 sentence overview of what the changes do and their impact]
## Security Review (External PR)
⚠️ This is an external PR from a fork. Additional security checks performed:
- [ ] No unauthorized access to secrets/env vars
- [ ] No suspicious network requests
- [ ] No hardcoded credentials
- [ ] No SQL injection risks
- [ ] Safe file system operations
[List any security concerns found or state "No security issues found"]
## Issues Found
Total: [X critical, Y important, Z minor]
### 🔴 Critical (Must Fix)
[Issues that will break functionality or cause data loss]
- **[Issue Title]** - `path/to/file.py:123`
Problem: [What's wrong]
Fix: [Specific solution]
### 🟡 Important (Should Fix)
[Issues that impact user experience or code maintainability]
- **[Issue Title]** - `path/to/file.tsx:45`
Problem: [What's wrong]
Fix: [Specific solution]
### 🟢 Minor (Consider)
[Nice-to-have improvements]
- **[Suggestion]** - `path/to/file.py:67`
[Brief description and why it would help]
## Performance Considerations
- Database query efficiency (no N+1 queries)
- Frontend bundle size impacts
- Async/await usage in Python
- React re-render optimization
[List any performance issues or state "No performance concerns"]
## Good Practices Observed
- [Highlight what was done well]
- [Patterns that should be replicated]
## Test Coverage
**Current Coverage:** [Estimate based on what you see]
**Missing Tests:**
1. **[Component/Function Name]**
- What to test: [Specific functionality]
- Why important: [Impact if it fails]
- Suggested test: [One sentence description]
## Recommendations
**Merge Decision:**
- [ ] Ready to merge as-is
- [ ] Requires fixes before merging
- [ ] Needs security review by maintainer
**Priority Actions:**
1. [Most important fix needed, if any]
2. [Second priority, if applicable]
**External PR Note:**
This PR is from a fork. Maintainers should manually verify the changes before merging.
---
*Review based on Archon V2 Alpha guidelines and secure external PR process*
unauthorized-message:
needs: check-authorization
if: needs.check-authorization.outputs.authorized == 'false' && needs.check-authorization.outputs.trigger_phrase == '@claude-review-ext'
runs-on: ubuntu-latest
permissions:
issues: write
pull-requests: write
steps:
- name: Post Unauthorized Message
uses: actions/github-script@v7
with:
script: |
const prNumber = '${{ needs.check-authorization.outputs.pr_number }}';
const commentAuthor = '${{ needs.check-authorization.outputs.comment_author }}';
const comment = {
owner: context.repo.owner,
repo: context.repo.repo,
issue_number: parseInt(prNumber),
body: `❌ @${commentAuthor} - You are not authorized to trigger Claude reviews.\n\n` +
`Only the following maintainers can trigger Claude: @Wirasm, @coleam00, @sean-eskerium\n\n` +
`Please ask a maintainer for review.`
};
await github.rest.issues.createComment(comment);