The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to read and execute instructions from external Implementation plans, which introduces a surface for indirect prompt injection if the plan source is untrusted.
Ingestion points: The agent is instructed to "Read plan file" in SKILL.md (Step 1).
Boundary markers: The skill includes strong instructions to "Review critically" and "Raise [concerns] with your human partner before starting."
Capability inventory: The skill allows for general task execution, which may include file system operations and command execution as defined in the plan.
Sanitization: Safety is managed through explicit human review checkpoints and mandatory stop conditions if instructions are unclear or blockers are encountered.

executing-plans