The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill functions as a retrieval-based Q&A system that ingests content from multiple local project files to answer user queries, which creates a potential surface for instructions embedded in those documents to influence agent behavior.
Ingestion points: The skill reads data from docs/USER_GUIDE.md, CHEATSHEET.md, knowledge/shared/GLOSSARY.md, README.md, CATALOG.md, and frontmatter from installed SKILL.md files.
Boundary markers: The instructions establish a strict 'contract' requiring the agent to only answer from the provided corpus, cite specific file:line references, and explicitly state when information cannot be found rather than hallucinating answers.
Capability inventory: The skill's execution environment is restricted to Read, Grep, and Glob tools, effectively preventing unauthorized file writes, network exfiltration, or arbitrary command execution.
Sanitization: The skill relies on source-referencing and a 'no-invent' policy to mitigate the risk of processing untrusted content from the documentation files.

fh