Autonomous crash analysis and bug fixing. Monitors crash reports from Cloudflare D1, deduplicates, analyzes with Codex 5.3 High, generates fixes, and creates...

This skill has clear inconsistencies you should resolve before installing: - Ask the author to explain and fix the mismatch between SKILL.md and index.js: SKILL.md says MiniMax M2.5 is available and 'no extra API key needed', but the code calls https://api.minimax.chat with Authorization: Bearer ${ZAI_API_KEY} (an env var not documented). It also mentions 'Codex 5.3 High' in the docs but the code does not call OpenAI. - Treat this as a potential data-exfiltration risk: the code sends crash details (including user_id and device info) to an external endpoint (minimax.chat). Only proceed if you trust that endpoint and the associated API key usage is documented and intended. - Never supply your personal GH_TOKEN with broad scopes. If you test, create a limited-service token with only the repo scopes needed and prefer a test repo. - Request an updated SKILL.md that lists all environment variables used (including ZAI_API_KEY) and documents where crash data is sent. If the skill must call an external AI API, ask for justification why it cannot use the platform-provided model. - If you cannot get clear answers, run the skill in a sandboxed environment (dry-run) with minimal secrets and monitor network egress to confirm where data is sent before granting production credentials.

Type: OpenClaw Skill Name: crash-fixer Version: 3.0.0 The skill is classified as suspicious due to its high-risk capabilities and a critical prompt injection vulnerability. It uses a GitHub token (`GH_TOKEN`) to read, create branches, commit AI-generated code, and create pull requests in a target repository (`TARGET_REPO`), effectively allowing it to modify source code. The `analyzeCrash` function in `index.js` constructs an AI prompt using raw, unsanitized crash data (e.g., `errorName`, `message`, `stackTrace`). A malicious actor could submit a crafted crash report containing prompt injection instructions, leading the AI to generate malicious `fix_code` or specify an arbitrary `file_path`, which the skill would then commit to the target repository. This represents a significant supply chain risk and potential for remote code execution via AI manipulation.