The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by ingesting data from untrusted markdown files.
Ingestion points: Processes files within the .scratch/ directory and tracker files such as .agents/issue-tracker.md.
Boundary markers: No specific delimiters or warnings to ignore instructions within the ingested data were identified.
Capability inventory: The skill coordinates several other pipeline tools including to-spec, to-tickets, and implement.
Sanitization: There is no logic for sanitizing or validating the contents of the ingested markdown files.

wayfinder