The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external issue trackers and project files to generate planning artifacts and register features on a roadmap.
Ingestion points: The skill reads project metadata from docs/features/ROADMAP.md and progress.md, and ingests external issue content via the --from-issue flag or numeric identifiers (e.g., #N).
Boundary markers: While the skill incorporates workflow validation gates (Redirect and Product-review gates), it does not define specific isolation delimiters (such as XML tags or unique string markers) to separate ingested issue descriptions from the agent's instructional context.
Capability inventory: The skill performs file reads and writes (generating SPECs, engineering artifacts, and roadmap updates) and interacts with the forge CLI to fetch external issue data.
Sanitization: There are no specific instructions for sanitizing, escaping, or filtering potential malicious instructions embedded in issue content before they are written to project documentation.

plan-feature