The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates deep exploration of a codebase, which involves ingesting content from arbitrary files.
Ingestion points: Files are read from the local repository using Read, Grep, and Glob tools as instructed in SKILL.md and referenced prompt templates.
Boundary markers: The skill does not define specific delimiters to separate code content from potentially malicious embedded instructions in the files being analyzed.
Capability inventory: The skill explicitly limits subagents to readonly: true in SKILL.md, preventing them from writing to the filesystem or executing destructive commands.
Sanitization: There is no mention of sanitizing file content before it is processed by the explaining and exploring agents.

how