The Agent Skills Directory

[SAFE]: The skill implements a structured, multi-step workflow for auditing developer surfaces. It includes clear activation rules and requires specific user intent before proceeding with file inspections or operations.
[PROMPT_INJECTION]: The skill exhibits a potential surface for indirect prompt injection (Category 8) because it ingests and processes untrusted data from the target project being audited (e.g., API documentation, CLI help strings, and SDK code).
Ingestion points: Target project files are loaded into context during the 'Load grounded context' step of the workflow.
Boundary markers: The templates use clear structural markers (Markdown headers) to separate findings from analysis.
Capability inventory: The skill writes report artifacts to the local filesystem (docs/audits/) and manages workflow state in JSON files.
Sanitization: While no explicit sanitization is mentioned, the risk is mitigated by the skill's primary purpose of code analysis and its reliance on predefined output templates. The severity is assessed as safe given it is fundamental to the skill's operation.
[EXTERNAL_DOWNLOADS]: The skill references a wide array of high-reputation industry sources, including guides and standards from Google, Microsoft, AWS, OpenAI, and Stripe. These are documented as grounding sources for the agent's heuristics and do not involve untrusted remote code execution.

dx-audit