The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell commands to discover project structure and perform audits. It uses find to identify manifests in SKILL.md and grep to search for credentials or patterns in references/universal-checks.md.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8). 1. Ingestion points: The agent reads the entire codebase, including source files and documentation, during the discovery process. 2. Boundary markers: There are no explicit delimiters or instructions to ignore content within ingested files. 3. Capability inventory: The agent can execute filesystem search commands and write architecture documentation. 4. Sanitization: No sanitization is performed on the ingested code or data before it is processed by the agent.
[SAFE]: The identification of sensitive files (like .env) and secrets is a documented auditing feature. No unauthorized network activity or persistence mechanisms were found.

project-blueprint