The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local utility scripts via Node.js to perform the audit's core logic, such as preflight checks, consolidating debate results, and formatting the final report. These executions are constrained to the skill's own internal script files and follow a secure workflow for temporary file management.
[DATA_EXFILTRATION]: To prevent the leakage of sensitive information, the skill includes a mandatory redaction phase. It uses specific patterns to identify and mask email addresses, blockchain addresses, and potential secrets (like private keys or API tokens) before providing the audit results to the user.
[PROMPT_INJECTION]: The skill is designed to ingest and process external specification documents, which presents a surface for indirect prompt injection. This risk is managed through a structured multi-phase process that isolates the audit logic in external scripts and implements mandatory output filtering via a redaction script. (Ingestion points: target file path; Boundary markers: relies on internal prompt templates; Capability inventory: shell execution, file writing, and skill invocation; Sanitization: output redaction filter).

necessity-audit