Monitor and clean invalid codex auth files from CPA (Codex Provider Agent). Checks quota status, disables 401 files, double-verifies before deletion. Use whe...

This skill is coherent in purpose — it contacts a Codex Provider Agent (CPA) management API and can disable or delete auth files — but there are important mismatches and risks you should consider before installing: - Missing metadata: The package metadata does not declare the required CPA_URL and CPA_KEY, yet the script needs them (via env vars or config.json). Expect to provide an admin key; ensure the registry omission isn't hiding additional secret requirements. - Destructive operations: The tool can permanently delete auth files on the CPA. Test with the status/check commands first and run against a staging instance. Back up or export current auth-file state before running delete/clean in production. - Principle of least privilege: Use a scoped admin key where possible. If the CPA supports a role that can only disable (not delete), prefer that for routine monitoring and only use a full-admin key when you intentionally run deletion. - Config file security: config.json will contain the admin key. Ensure the file is stored with restrictive permissions and verify where it will be saved (the skill directory under ~/.nanobot/workspace in examples). - Notifications unclear: The config includes Telegram/Discord settings but does not show where a bot token or webhook is supplied. Inspect the remainder of the script to confirm how notifications are sent and whether any additional secrets are required. - Review & auditing: Because the script performs network calls and deletions, review the full script (including the truncated portion) to confirm there are no unexpected external endpoints, no hidden logging/exfiltration, and that the delete logic matches your expectations (e.g., only deletes items confirmed by the double verification). If you want to proceed safely: (1) run the script in 'status' and 'check' modes against a test CPA instance, (2) inspect the complete source to confirm notification code and external endpoints, (3) restrict the CPA key used, and (4) ensure secure storage/rotation of the admin key. If you can provide the remainder of the script (truncated portion), I can re-evaluate and reduce uncertainty.

Type: OpenClaw Skill Name: codex-cleaner Version: 1.0.3 The skill is designed to monitor and clean codex authentication files via a CPA management API, which aligns with its stated purpose. It uses standard Python libraries, handles an admin API key (CPA_KEY) for authentication, and stores configuration in a local `config.json` file. While it interacts with a `/v0/management/api-call` endpoint, the target URL for this call is hardcoded to `https://chatgpt.com/backend-api/wham/usage`, preventing its use for arbitrary SSRF or malicious proxying from within this skill. There is no evidence of intentional data exfiltration, persistence mechanisms, or prompt injection attempts against the agent in `SKILL.md`.