The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions involve reading and processing "upstream agent specs" and "prompt files", which are untrusted external inputs.
Ingestion points: Workflow Step 2 in SKILL.md requires the agent to "Read the source material" from upstream documents or files to extract roles, workflows, and constraints.
Boundary markers: The instructions do not define boundary markers or "ignore embedded instructions" directives to prevent the agent from being influenced by directives potentially hidden within the source material.
Capability inventory: The skill allows the agent to generate or rewrite text for SKILL.md, agents/openai.yaml, and README.md files.
Sanitization: No specific sanitization, escaping, or filtering is prescribed for the source content beyond general rewriting for clarity and repository fit.

skill-converter