The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's core workflow involves reading and analyzing untrusted external data from the project repository to generate clarification questions. \n
Ingestion points: As described in SKILL.md and the contract reference, the skill reads files such as README.md, AGENTS.md, docs/adr/, and various source code files. \n
Boundary markers: The skill requires the agent to maintain a clear separation between verified facts, user decisions, and the question frontier, which helps mitigate accidental instruction following from ingested data. \n
Capability inventory: Logic is delegated to specialized capabilities like socratic and light-research rather than using raw system commands or execution sinks. \n
Sanitization: The instructions do not specify any explicit validation or sanitization of the content read from local project files.

project-clarify