MessageGuard

Filters outgoing text for sensitive data using regex patterns and configurable actions like masking, blocking, or warning to prevent secret leaks.

What to consider before installing/using MessageGuard: - The skill is coherent: it implements a regex-based outgoing message filter as described. There are no network exfiltration or hidden installers in the provided files. - Logging risk: if you enable log_detections, detection events (including snippets) are written to disk (default ~/.openclaw/outgoing-filter.jsonl). That file may contain sensitive data. If you enable logging, choose a secure path and file permissions or keep logging disabled. - Allow-listing bypass: the allow_channels config accepts channel IDs that completely bypass filtering. Only add trusted, secure channels to that list. - Verify integration: callers must check the script's exit code and handle blocked responses correctly; otherwise blocked content could be accidentally sent. Test the wrapper behavior in your environment before relying on it. - Review patterns: the built-in pattern library is aggressive (blocks many key formats). Review & tune patterns to reduce false positives (or disable patterns you don't want). - Source provenance: SKILL.md suggests cloning from an SSH GitHub repo but the bundle contains the code. Because the registry 'Source' is unknown and there is no homepage, if provenance matters for your environment, inspect the included script and configs locally or run in an isolated environment. Consider auditing the code and config files before enabling in production. - Operational advice: keep default log_detections=false unless you need audits; if you enable logs, secure the log file. Use conservative allow_channels and test with sample messages to confirm behavior. If you want, I can: (1) point out specific lines in the Python script that write logs or read config files, (2) produce a short checklist to harden configuration, or (3) highlight patterns you might want to disable for your use case.

Type: OpenClaw Skill Name: messageguard Version: 0.1.0 The OpenClaw AgentSkills 'MessageGuard' skill is designed to filter outgoing messages for sensitive information, acting as a security control. The Python script `scripts/filter_message.py` implements this by using a comprehensive set of regex patterns to detect and then mask, block, or warn about sensitive data like API keys, private keys, and PII. All dependencies are from the Python standard library, and the code is clear and well-documented. While there are theoretical configuration-based vulnerabilities (e.g., ReDoS if a malicious regex is supplied in config, or arbitrary file write if `log_path` is maliciously controlled), these are not indicative of intentional malice within the skill's design. The skill's explicit purpose and implementation are to enhance security, not to compromise it.