The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Bash, Grep, and Glob to examine local project files and configurations. This behavior is necessary for performing the security audit tasks described in the instructions.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it reads and processes untrusted data from local files (e.g., source code, environment files) that could contain malicious instructions designed to subvert the auditing process.
Ingestion points: Files like package.json, .env, and source code are read using the Read and Grep tools.
Boundary markers: The skill does not implement specific delimiters or instructions to ignore potential commands embedded within the files being audited.
Capability inventory: The agent has access to Bash, Read, WebFetch, and WebSearch tools.
Sanitization: There is no mention of sanitizing or validating the content of project files before processing them.
[DATA_EXFILTRATION]: By design, this skill searches for and collects sensitive information such as API keys and credentials to report them as vulnerabilities. Users should be aware that these secrets are included in the tool's execution context and reports. The delivery instructions mention using a command /atlas-report to handle detailed findings.

warden-audit