The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents a framework for task chaining that interpolates untrusted metadata into follow-up actions. * Ingestion points: Frontmatter hooks definitions in SKILL.md (e.g., prd_complete, implementation_complete). * Boundary markers: None defined in the hook format or implementation guide to delimit interpolated data. * Capability inventory: The triggered skills listed in examples (e.g., prd-planner, create-pr) possess capabilities like Read, Write, Edit, and Bash. * Sanitization: The implementation guide lacks instructions for sanitizing or escaping the interpolated context variables such as {feature_name} or {prd_file} before they are passed to the next agent session.

auto-trigger