The Agent Skills Directory

[SAFE]: The skill focuses on validating product-market fit by reading a local PRODUCT.md file and generating summary artifacts in a local .agents/design/ directory. All operations are consistent with the documented purpose of a development workflow tool.
[COMMAND_EXECUTION]: Uses benign shell commands (ls, mkdir) for directory management and file existence checks. The provided validation script (scripts/validate.sh) performs static checks on the skill's own structure using grep and file tests, posing no security risk.
[PROMPT_INJECTION]: The skill has an indirect injection surface as it ingests content from PRODUCT.md to inform its alignment matrix and the downstream council skill. While this content is untrusted, the skill uses it within a structured rubric framework, which mitigates simple adversarial influence.
Ingestion points: PRODUCT.md (Step 1).
Boundary markers: Absent; content is interpolated into scoring rationales.
Capability inventory: Local file system access (ls, mkdir, file write) and invocation of the council skill.
Sanitization: No explicit filtering of the PRODUCT.md content before processing.

design