Safe Guardian

A safe security layer plugin for OpenClaw that intercepts dangerous tool calls (exec, write, edit) through multi-layer blacklist/whitelist filtering and inte...

This package implements local blacklist/whitelist checks and logging, but there are several red flags you should consider before installing or enabling it: - Documentation mismatch: SKILL.md repeatedly describes LLM-based intent validation and shows an apiKey field, but the shipped code performs only local regex checks and does not use any API keys or network calls. Treat LLM-related claims as unimplemented until the author updates the code. - API mismatch: Examples in the README call checkToolCall with a tool type and options object, while the actual implementation expects a single string. Your integrations may break if you rely on the documented API. - Silent overwrite risk: Initializing SafeGuardian will unconditionally write ./config/blacklist.json (createDefaultBlacklist), overwriting any existing blacklist file in that path. Run it in a sandbox first — do not run from a directory containing important configs or as root. - File writes and logs: The module creates ./config and ./logs/guardian and appends to a log file. Ensure the working directory is safe and that log/config locations are acceptable for your environment. - Testing recommendations: Run the package in an isolated environment (container or VM) and review behavior: confirm whether it overwrites configs, verify API semantics, and ensure it won’t be invoked with sensitive working directories. If you expect LLM-based intent checks, request that the author implement and document secure LLM integration (network calls, explicit env var requirements, and opt-in behavior) before trusting the skill for production use. Given these inconsistencies, do not deploy this skill to production or give it autonomous privileges until the author resolves the documentation/implementation mismatches and provides safer initialization behavior (e.g., only create default config when no config file exists, or prompt/backup existing config).

Type: OpenClaw Skill Name: safe-guardian Version: 1.1.1 The 'safe-guardian' skill is a security utility designed to intercept and filter potentially dangerous tool calls (like shell execution or file modifications) using blacklists, whitelists, and audit logging. The implementation in index.js and the configurations in config/blacklist.json and config/whitelist.json are consistent with the stated purpose in SKILL.md, and no evidence of data exfiltration, malicious execution, or prompt injection was found.