The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses uvx skilllint@latest to perform validation during the refactoring process. This command fetches the skilllint package from a remote registry at runtime.\n- [REMOTE_CODE_EXECUTION]: By using the uvx tool, the skill executes code downloaded from an external source (the skilllint package). This execution is used for the primary purpose of validating skill structure and token counts.\n- [COMMAND_EXECUTION]: The workflow involves executing local shell commands including grep to search for references and uvx to run the linter.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external files (the skills to be refactored) provided at the $ARGUMENTS path. This creates a surface for indirect prompt injection if the processed content contains instructions that influence the agent's refactoring logic. The skill lacks explicit sanitization or boundary markers for this external content.\n
Ingestion points: The contents of the skill directory provided via $ARGUMENTS in the Phase 1 Domain Analysis step.\n
Boundary markers: None identified in the workflow instructions.\n
Capability inventory: File writing (SKILL.md), file moving (references), and command execution (uvx, grep).\n
Sanitization: No explicit sanitization or escaping of external skill content is described.

refactor-skill