The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for indirect prompt injection by reading external, potentially untrusted content.
Ingestion points: SKILL.md (Step 1: Instructions to read existing docs and code comments).
Boundary markers: Absent; no delimiters or instructions to ignore embedded commands are specified to isolate the external content from the agent's instructions.
Capability inventory: SKILL.md (The frontmatter does not limit allowed-tools, meaning the agent retains access to its full default suite of tools while processing untrusted content).
Sanitization: Absent; the skill does not include any requirements for validation or filtering of the ingested data.

docs