The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides a retrieval-augmented documentation system that can ingest and display instructions embedded in project files.
Ingestion points: The brief command in runtime_src/40_retrieval.py reads user-provided task descriptions and retrieves excerpts from knowledge cards and documents within the .codestable/ directory.
Boundary markers: The retrieved knowledge is presented in the prompt using standard Markdown formatting (headers and blockquotes) but lacks explicit delimiters or instructions to ignore embedded commands.
Capability inventory: The tool possesses several capabilities that could be leveraged if an injection occurs, including executing Git commands for drift and audit checks, writing Markdown files to the local repository, and modifying the project's .codestable/config.json via the learn and topics update commands.
Sanitization: While the skill includes a specialized detect_secret function in runtime_src/00_core.py to prevent the leakage of credentials and private keys into the knowledge base, it does not provide sanitization to filter out malicious natural language instructions from retrieved content.

cs