The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted external data files to generate technical specifications and implementation plans for other agents.
Ingestion points: The skill reads multiple external files including feature_list.json, AGENTS.md, PROGRESS.md, CONTEXT.md, DESIGN.md, and various product/discovery documents.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to disregard malicious commands or override attempts that might be embedded within the ingested data files.
Capability inventory: The skill has the capability to read any file in the repository for research purposes and write output to docs/specs/<feature-id>.md.
Sanitization: The instructions do not mention any sanitization, validation, or filtering of the content retrieved from external files before it is interpolated into the generated specification. A malicious entry in feature_list.json (e.g., in the user_visible_behavior or verification fields) could influence the resulting technical plan, potentially tricking a downstream implementation agent into performing unsafe actions.

feature-spec