feat: PR security checks — suspicious paths, committer identity, auto-merge override

[myakove]

Summary

Add security checks to detect and block malicious PR attack vectors, inspired by recent GitHub supply-chain attacks where attackers hide payloads in collapsed diffs, inject trojan configs (.claude/, .vscode/), and tamper with automated PRs.

Problem

GitHub's UI collapses large diffs and generated files, allowing attackers to hide malicious code in PRs that appear clean at a glance. Additional attack vectors include:

1. Trojan config directories — .claude/settings.json with "command": "node .claude/setup.mjs" executes malware on clone
2. CI/CD pipeline tampering — modifications to .github/workflows/ can exfiltrate secrets
3. PR tampering — attacker pushes commits on top of a bot's PR, changing the last committer identity

The webhook server already clones the repo and sees the full diff — no GitHub UI collapsing applies. We can inspect everything.

Proposed Solution

Three configurable security checks, all working at all 3 config levels (global config.yaml → per-repo in config.yaml → in-repo .github-webhook-server.yaml):

1. Suspicious Path Detection (new check run: security-suspicious-paths)

• Configurable list of path prefixes under security-checks.suspicious-paths
• Default blocked paths:
  • .claude/ — AI agent trojan configs
  • .vscode/ — editor config injection
  • .cursor/ — AI editor config injection
  • .devcontainer/ — container config injection
  • .pi/ — pi agent config injection
  • .github/workflows/ — CI/CD pipeline tampering
  • .github/actions/ — CI/CD action tampering
• When a PR changes files matching any pattern → fail the check run with details of which files matched
• Users can extend or override the list at any config level

2. Committer Identity Check (new check run: security-committer-identity)

• Compare last_committer vs parent_committer (PR author)
• When they differ → fail the check run with a warning identifying the unexpected committer
• Detects the MintMaker-style tampering pattern where an attacker pushes commits on top of a bot's automated PR
• Configurable: security-checks.committer-identity-check: true (default: true)

3. Auto-Merge Override

• In pull_request_handler.py auto-merge logic, after auto_merge = True is determined
• Check changed files against suspicious-paths list
• If any match → set auto_merge = False, post comment: "Auto-merge blocked: PR modifies security-sensitive paths: [list of matched files]"
• No new check run — just blocks the auto-merge decision
• Ensures PRs touching sensitive paths are NEVER auto-merged, even for auto_verified_and_merged_users

Configuration

security-checks:
  suspicious-paths:                # list of path prefixes to flag
    - ".claude/"
    - ".vscode/"
    - ".cursor/"
    - ".devcontainer/"
    - ".pi/"
    - ".github/workflows/"
    - ".github/actions/"
  committer-identity-check: true   # flag when last committer ≠ PR author (default: true)

Configurable at all 3 levels:

1. Global — config.yaml (server-level defaults for all repos)
2. Per-repo — config.yaml under repositories: section
3. In-repo codebase — .github-webhook-server.yaml in the repository itself

Files to Modify

• webhook_server/config/schema.yaml — add security-checks schema
• webhook_server/libs/github_api.py — read security-checks config in _repo_data_from_config()
• webhook_server/libs/handlers/runner_handler.py — new run_security_checks() method
• webhook_server/libs/handlers/pull_request_handler.py — integrate security checks in PR processing + auto-merge override
• webhook_server/utils/constants.py — new check run name constants
• webhook_server/tests/ — tests for all 3 checks
• examples/.github-webhook-server.yaml — example config

Done

• Add security-checks to config schema with suspicious-paths and committer-identity-check
• Read security config in _repo_data_from_config()
• Implement suspicious path detection check run
• Implement committer identity check run
• Implement auto-merge override for suspicious paths
• Add security checks to PR processing pipeline
• Add to welcome message (inform users which security checks are active)
• Tests for suspicious path detection
• Tests for committer identity check
• Tests for auto-merge override
• Update example config