The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill performs audits by reading and interpreting data from AI-SPEC.md and historical phase outputs. These files are untrusted inputs that could contain instructions designed to influence the agent's behavior.
Ingestion points: Accesses AI-SPEC.md and outputs from completed AI phases (referenced in the objective and context).
Boundary markers: The skill does not define delimiters or specific instructions to isolate or ignore embedded prompts within the documents it processes.
Capability inventory: The skill utilizes Read, Write, Bash, and Agent tools to execute its audit logic.
Sanitization: No sanitization, validation, or filtering of the content read from external files is described.

gsd-eval-review