The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and analyzes technical documents provided by users. These documents are untrusted data sources that could potentially contain instructions designed to override the agent's briefing logic.
Ingestion points: Technical markdown documents are read via the <doc-path> parameter defined in SKILL.md.
Boundary markers: The skill employs structured extraction heuristics and specific output templates (e.g., Assumptions Register, Reasoning Chain) to focus the agent on specific technical data points, reducing the impact of unstructured natural language instructions.
Capability inventory: The skill utilizes Read, Grep, Glob, Write, and Bash tools to perform its analysis and verification tasks.
Sanitization: The skill includes a robust redaction scan in references/extraction-guide.md that automatically aborts if high-confidence secrets (e.g., AWS keys, GitHub tokens) are detected and masks medium-confidence secrets.
[COMMAND_EXECUTION]: The verification workflow triggers an independent reasoning audit that executes shell commands including git status, git diff, ls, cat, and grep. These tools are used to validate document claims against the actual state of the project. The skill enforces repository boundary checks and path normalization to ensure these commands do not access files outside the intended project scope.

fp-brief