The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests untrusted data from external documentation sources to generate files and instructions.
Ingestion points: The reference-researcher and skill-coverage-analyzer sub-agents fetch data from external URLs (either provided via the base_url argument or discovered via web search) to understand the target library.
Boundary markers: The instructions do not define boundary markers or provide specific "ignore embedded instructions" warnings to the sub-agents when processing external web content.
Capability inventory: The skill has the capability to write multiple files to the local file system (under skills/<library>/), update the root CLAUDE.md file, and execute the create-commit skill to persist changes to version control.
Sanitization: There is no mention of sanitizing, filtering, or validating the content retrieved from external sources before it is passed to the skill-author or other agents to generate the final skill instructions and scripts.

create-skill