Security audit
clawtrial
Security checks for vulnerabilities and agentic risk
Overview
The package superficially matches an 'agent monitoring' purpose but contains multiple instructions and install-time behaviors that modify agent system prompts, add automatic silent execution hooks, and enable external case submission — these behaviors are disproportionate and risky without careful review.
This package is not clearly malicious but contains several risky behaviors you should evaluate before installing. If you consider installing: - Do NOT accept default setup blindly. Inspect scripts/postinstall.js, scripts/setup-cron.js and any postinstall actions; run them in a sandbox first (VM/container) rather than on your primary machine. - The package attempts to modify agent system prompts and append 'auto-execution' instructions so your agent will act silently. If you value control, refuse or remove those modifications (do not add the AGENT_CONFIG entries or SOUL.md/AGENTS.md edits). - The default configuration enables an external API endpoint. If you want strictly local operation, set api.enabled = false in the saved config before allowing network activity, or block the package's network access until you audit src/api.js for what it sends. - The postinstall will create symlinks in agent skill/plugin directories and write config files and keys. Review and, if necessary, perform linking manually rather than allowing an automated postinstall. - If you lack the ability to audit the code: run the package only in an isolated environment, or decline installation. Ask the maintainer for an authoritative source (official homepage/repository) and a clear security/privacy policy. If the package was installed from untrusted sources or the repo doesn't match the package metadata, do not install.
SkillSpector findings are pending for this release.