Openclaw Warden

Verify workspace file integrity and scan for prompt injection patterns in agent identity and memory files. Detects unauthorized modifications to SOUL.md, AGENTS.md, IDENTITY.md, memory files, and installed skills. Free detection layer — upgrade to openclaw-warden-pro for automated countermeasures.

This skill is coherent with its stated purpose and runs entirely locally under python3, but it will create .integrity state in your workspace and can rename/quarantine other skill directories and restore files. Before installing or enabling autonomous invocation: (1) inspect scripts/integrity.py (especially quarantine/restore/protect/rollback implementations) to confirm their behavior matches your expectations, (2) back up your workspace, (3) consider running it manually first rather than granting automated startup hooks or autonomous agent invocation, and (4) verify there are no network calls or hidden endpoints in the remainder of the code if you will run it with sensitive data.

Type: OpenClaw Skill Name: openclaw-warden Version: 1.0.3 The OpenClaw Warden skill is a security tool designed to detect unauthorized modifications and prompt injection attacks within an AI agent's workspace. The `SKILL.md` and `README.md` clearly state its defensive purpose, and the `scripts/integrity.py` code implements integrity checks (SHA-256 hashing), injection pattern scanning (for instruction overrides, base64 payloads, exfiltration URLs, shell injection, etc.), and countermeasures like file restoration from local snapshots or Git, and skill quarantine. The code uses only Python standard library modules, makes no network calls, and all operations are local file system or Git commands, confirming its stated 'no external dependencies' claim. There is no evidence of malicious intent, data exfiltration, backdoors, or prompt injection against the agent itself; rather, it actively defends against these threats.