The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes instructions from externally provided markdown specification files (e.g., docs/...-spec.md) to drive its behavior and tool usage.
Ingestion points: The skill reads a user-provided path to a specification file in the Entry section of SKILL.md.
Capability inventory: The skill is capable of invoking other skills (context-init, context-checkpoint, and dynamic skills named within the spec), reading files, and writing to the specification file to track progress.
Boundary markers: There are no explicit delimiters or instructions to the agent to treat the content of the specification file as untrusted data rather than direct instructions.
Sanitization: The skill does not perform sanitization or validation on the strategies or skill names retrieved from the specification file, relying instead on a manual 'Critical Review' step by the agent before execution begins.

executing-specs