The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill is designed to ingest and analyze external product documentation, roadmaps, and planning docs. This data ingestion path creates a surface where instructions embedded within the analyzed content could potentially influence the behavior of the evaluation personas.
Ingestion points: The skill accepts file paths, repository directories, or text descriptions through the $ARGUMENTS parameter in SKILL.md.
Capability inventory: It utilizes the delegate tool to spawn independent sub-sessions for six different review lenses.
Boundary markers: The skill uses structured templates to pass information to the lenses, though it does not explicitly define strict sanitization or escaping for the interpolated external content.
Sanitization: No explicit sanitization of the input documents is mentioned before processing.
[External Skill Dependencies]: The skill orchestration depends on external components, such as the outcomist persona (expected at ~/.amplifier/skills/outcomist) and the amplifier-bundle-skills from the author's organization. These references are used to load the necessary analysis lenses for the council.
[Isolated Execution Design]: The skill includes a proactive guard check to ensure it only processes explicit external targets when running in a forked session, preventing it from incorrectly accessing or fabricating responses based on local conversation history it cannot see.

product-council