The Agent Skills Directory

[SAFE]: No malicious patterns or security vulnerabilities were detected in the skill instructions or rule files.
[COMMAND_EXECUTION]: The skill performs local command execution to audit CLI tools (e.g., node ./dist/cli.js). The instructions explicitly limit this to safe probe sets such as --help and --version to avoid destructive side effects.
[DATA_EXFILTRATION]: The skill reads local repository files (package.json, manifests, source code) to perform its audit. This access is restricted to the public surface of the project being audited and does not involve exfiltration of sensitive data.
[PROMPT_INJECTION]: While the skill contains instructions to rank findings as CRITICAL or HIGH, this is a reporting convention for DX issues and not an attempt to override the AI's safety guidelines or core behavior.

dx-audit