github-workflow-automation

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 1.00). The workflows repeatedly send repository diffs and full file contents to external AI services and include automated git operations that can force-push or run arbitrary exec commands — creating clear data-exfiltration and repository-compromise risks.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.85). In the required workflow for “Automated PR Review / On-Demand Assistance”, the agent’s LLM input includes outsider-authored PR diff/issue/comment body text (e.g., from pull_request code changes and issue_comment/pull_request_review_comment), which is read at runtime via ${{ steps.diff.outputs.diff }} and context.txt (cat context.txt) and then interpolated into the AI prompt/ai.chat(...).

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.70). The workflows declare GitHub Action repositories that are fetched and executed at runtime (e.g., actions/github-script@v7, actions/checkout@v4, actions/stale@v9, slackapi/slack-github-action@v1), so external code is pulled and run as required dependencies.