The Agent Skills Directory

[COMMAND_EXECUTION]: Uses the find command to resolve local plugin paths and grep to scan project configuration files for specific technology patterns.
[DATA_EXPOSURE]: Reads package.json and CSS configuration files to gain context about the project's tech stack.
[SAFE]: Enforces human-in-the-loop validation for the 'interaction thesis' and strictly forbids dependency installation without explicit permission.
[PROMPT_INJECTION]: The skill ingests untrusted project metadata which creates an indirect prompt injection surface.
Ingestion points: package.json, tailwind.config.*, and postcss.config.* files.
Boundary markers: None used around data ingestion.
Capability inventory: Full Bash, Read, Write, Edit, and WebSearch tools.
Sanitization: Employs specific regex filtering via grep to isolate known library names, which significantly reduces the risk of interpreting arbitrary file content as instructions.

creative-excellence