The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process user-provided UI code (React components, hooks, etc.) to perform refactoring. This creates an inherent surface for indirect prompt injection, where malicious instructions could be embedded within code comments or documentation of the target files. The skill mitigates this risk by requiring a 'preview plan' step where the agent must describe intended changes before execution, allowing for human review. It also defines strict behavior invariants to prevent unexpected logic changes.
Ingestion points: Target source code files provided by the user for refactoring (SKILL.md).
Boundary markers: No explicit content delimiters are defined, but the workflow enforces a 'preview first' gate (SKILL.md).
Capability inventory: File write/edit capabilities for code modification and command execution for running characterization tests (SKILL.md, references/refactor-passes.md).
Sanitization: No explicit sanitization or filtering of input code content is mentioned beyond logic-preserving checks.
[COMMAND_EXECUTION]: The instructions guide the agent to use local CLI tools such as omh, claude, and codex to record runtime evidence and manage delegation. It specifically suggests granting permissions like --permission-mode acceptEdits to subagents. While this automates file modifications, the skill mandates that this only occurs after a manual preview of the refactor plan, maintaining user oversight.

omh-frontend-refactor