The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from external sources and user sessions, creating a potential attack surface.
Ingestion points: The skill reads project posture from docs/agents/project.md and anchor subjects from user-provided pastes.
Boundary markers: The skill implements 'Iron Laws' (e.g., 'CODEBASE TRUTH BEFORE PASTE') and a mandatory 'Claim hygiene' output block (Source/Verified/Inference/Open) to distinguish between trusted repository facts and untrusted external inputs.
Capability inventory: The skill contains no scripts, code files, or network operation tools; it relies entirely on instructions for the agent's reasoning.
Sanitization: The skill uses instructional logic, such as mandatory re-derivation of knowledge maps for every new subject, to prevent instructions or context from leaking between different analysis domains.
[NO_CODE]: The skill consists entirely of Markdown and YAML configuration files and does not include any executable scripts or binary files.

deepen-codebase