Security audit
Security checks for vulnerabilities and agentic risk
This skill is a repository review helper that reads project files and writes a local JSON report, with its behavior generally disclosed and aligned to its purpose.
Before installing, expect the skill to read repository source files, optionally scan the whole project when requested with --all, use local CLI tools, and write a .beagle/llm-artifacts-review.json report into the working tree. Review that generated file before committing it.
The manifest description says the skill triggers on 'dead code sweeps,' 'test-quality passes,' and 'when the user asks to scan the whole repo.' These are broad natural-language conditions rather than a specific invocation list, which increases the chance of unintended activation during ordinary repository review conversations.
The skill is presented as a review/detection tool, but it also persists a JSON artifact inside the repository workspace. Writing files changes the working tree and can have unintended side effects in CI, pre-commit workflows, or downstream automation that trusts repo contents, so this exceeds a strictly read-only review capability.
The prose claims a historical staleness check against a previous report, but the actual shell snippet reads .beagle/llm-artifacts-review.json after Step 5 has written the current report and then compares that stored head to the current one. That logic does not implement the documented 'previous report exists' behavior and will not meaningfully detect report staleness as described.
The declared purpose is artifact review, but the implementation depends on multiple local command executions for repository inspection, directory creation, JSON parsing, and integrity checks. While some local file inspection is expected, requiring command execution and environment tooling is a broader capability than the manifest explicitly communicates.
No suspicious patterns detected.
