The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted repository source code and instructions during tasks like bug fixes and feature development.
Ingestion points: Repository files and documentation located in the user's workspace.
Boundary markers: The skill advises parsing and validating at system boundaries but lacks strict technical enforcement between data and instructions.
Capability inventory: Includes file system modifications, git operations, and execution of test suites or benchmarks.
Sanitization: Instructions include removing filler and host jargon (via 'unslop') from prose outputs.
[DYNAMIC_EXECUTION]: The skill dynamically references and reads instructions from other skills located in sibling directories.
Evidence: The SKILL.md instructions direct the agent to 'read its full SKILL.md at ..//SKILL.md' based on the principle index provided in the skill metadata.
[EXTERNAL_DOWNLOADS]: The skill references source material from a public repository for attribution purposes.
Evidence: THIRD_PARTY_NOTICES.md cites code adapted from the official Cursor plugins repository on GitHub.

engineering-mode