The Agent Skills Directory

Indirect Prompt Injection Surface: The skill processes untrusted content from user prompts and project metadata to generate a clarification.md artifact which is then consumed by downstream agents.
Ingestion points: Untrusted data enters the agent context through the userInput and project_facts inputs defined in SKILL.md.
Boundary markers: The instructions for generating clarification.md in templates/clarification.md do not explicitly mandate the use of delimiters or specific instructions to downstream agents to ignore potential commands embedded within the summarized user content.
Capability inventory: The skill is designed to write a persistent artifact to the filesystem ({{BASE_PATH}}/clarification.md), as outlined in the Outputs section of SKILL.md.
Sanitization: The workflow described in SKILL.md (Step 5 and 6) lacks explicit requirements to sanitize or escape user-provided strings before they are interpolated into the YAML front-matter or the body of the output Markdown file.

clarifying-scenarios