ragflow-runbook

End-to-end runbook for deploying, operating, troubleshooting, and monitoring RAGFlow (runtime ops only).

This skill appears to do what it says: an ops runbook for RAGFlow. Before installing/using it: 1) Keep your RAGFLOW_API_KEY secret (export it in your shell or a secrets manager; do not paste into chat). 2) Inspect scripts (especially deploy.sh) before running; the fallback download path fetches files from raw.githubusercontent.com but only if you set RAGFLOW_RUNBOOK_ALLOW_DOWNLOAD=1. 3) Starting containers is disabled by default; set RAGFLOW_RUNBOOK_ALLOW_START=1 only when you intend to run docker compose up. 4) The alert helper calls the openclaw CLI and requires an OpenClaw chat target if used — install/configure OpenClaw first or avoid alerting. 5) If you need extra assurance, run these scripts in an isolated/test environment (not production) the first time. The only minor inconsistency is that registry metadata declared no required env vars while SKILL.md does — this is documented in the changelog as an intentional duplication for scanners and does not materially change the skill's behavior.

Type: OpenClaw Skill Name: ragflow-runbook Version: 0.1.4 The skill bundle is designed for RAGFlow runtime operations, including deployment, health checks, and alerting. All scripts and documentation align with this stated purpose. Notably, `scripts/deploy.sh` includes explicit opt-in mechanisms (`RAGFLOW_RUNBOOK_ALLOW_DOWNLOAD`, `RAGFLOW_RUNBOOK_ALLOW_START`) to prevent unauthorized downloading of upstream files or starting containers, which is a strong security safeguard. The `SKILL.md` and other documentation provide clear instructions and security warnings (e.g., regarding API key handling and default passwords), without any evidence of prompt injection attempts to manipulate the agent maliciously. Python scripts (`ragflow_ping.py`, `ragflow_smoke.py`, `ragflow_status.py`) perform standard API calls for health monitoring, and `ragflow_alert.py` uses the `openclaw message send` CLI for intended operational alerts, explicitly warning against including secrets. There is no evidence of data exfiltration, unauthorized persistence, or malicious execution.