The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local shell scripts and an embedded Python script for output validation.
Evidence: scripts/validate.sh and scripts/validate-output.sh perform integrity checks on the skill's own structure and produced JSON files using standard utilities like grep and the Python json module.
[DATA_EXPOSURE]: The skill requires access to repository facts and intent sources to perform its review function.
Evidence: SKILL.md workflow steps 1 and 3 outline the inspection of acceptance criteria, non-goals, and repository patterns. This behavior is consistent with the primary purpose of a design-review skill and does not involve exfiltration.
[PROMPT_INJECTION]: The instructions incorporate a derivation-diff methodology to ensure independent judgment.
Evidence: The 'Derivation-diff challenge' section in SKILL.md explicitly instructs the agent to derive designs independently from ground truth, which serves as a defense against bias or instructions embedded within the author's initial plan.
[COMMAND_EXECUTION]: The skill instructs the agent to simulate defeat attempts but imposes strict operational boundaries.
Evidence: The 'Boundary' section in SKILL.md prohibits the agent from implementing, operating Git, or releasing work, ensuring that hypothetical failure constructions do not result in actual system modifications or unauthorized command execution.

premortem