The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted data from the conversation history and user intent to generate instructional content (SKILL.md) and executable resources.
Ingestion points: references/workflow.md mentions extracting context from conversation history (tools used, sequence of steps) and capturing user intent.
Boundary markers: The instructions do not define delimiters or explicit warnings to ignore embedded instructions within the processed data.
Capability inventory: The skill workflow includes file system writes (SKILL.md, bundled scripts), spawning subagents for testing, and executing scripts for automated grading.
Sanitization: No specific sanitization or filtering logic is prescribed for the ingested content before it is interpolated into the generated skill files.
[DYNAMIC_EXECUTION]: The workflow documentation encourages the agent to generate and run code dynamically to support the skill development process.
Evidence: references/workflow.md (Step 4) instructs the agent to "write and run a script rather than eyeballing" for grading runs, and the "Improve the skill" section suggests bundling repeated work into scripts/ to be used by the skill at runtime.

skill-creator