The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to execute the npx skills add command during the verification phase to ensure the polished skill is correctly registered.
[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection because it reads and processes external SKILL.md files which may contain untrusted instructions.
Ingestion points: The 'Intake' step (defined in SKILL.md and references/workflow.md) reads existing manifest files from the local directory.
Boundary markers: There are no explicit markers or 'ignore' instructions specified to encapsulate the content of the files being processed.
Capability inventory: The skill has file read and write permissions for refactoring tasks and shell command execution capabilities.
Sanitization: The instructions do not define any validation or sanitization procedures for the content extracted from target manifest files.
[EXTERNAL_DOWNLOADS]: The use of npx skills involves fetching and executing a package from the npm registry if it is not already available locally.

polish-skill-md