The Agent Skills Directory

[PROMPT_INJECTION]: The skill establishes context by reading untrusted repository files, creating a potential vector for indirect prompt injection.
Ingestion points: Files such as CONTEXT.md, docs/specs/INDEX.md, docs/agents/project.md, and docs/product/vision.md are read directly from the repository.
Boundary markers: The instructions do not define delimiters or specific warnings to ignore instructions that might be embedded in these project files.
Capability inventory: The skill uses tools like grep, writes to .skills/*.md and docs/roadmap/INDEX.md, and has the capability to trigger further skills like tdd or write-requirements.
Sanitization: There are no mentioned mechanisms for sanitizing or validating the content extracted from repository documents before use.

brainstorm