The Agent Skills Directory

[SAFE]: No security issues were detected. The skill follows a structured consensus model to validate planning decisions before handing off to execution agents, which serves as a design-level safety control.
[COMMAND_EXECUTION]: The skill references standard developer tools such as git, npm, and gh, as well as project-specific utilities like omx. These are used for repository exploration and task management in line with the skill's primary purpose.
[DATA_EXPOSURE]: The skill manages task context within the .omx/context/ directory. This is used for session persistence and grounding, involving only task-related facts and codebase touchpoints rather than sensitive system-level credentials.
[PROMPT_INJECTION]: The skill defines a 'Pre-Execution Gate' that interprets specific user prefixes like force: or ! as bypass signals for the planning workflow. This is a functional design choice for user experience rather than a malicious instruction override.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data from the codebase and user prompts, creating an attack surface for indirect injection.
Ingestion points: User-provided task descriptions, codebase content via omx explore, and external documentation via researcher.
Boundary markers: Employs structured markdown headers and a specific 'RALPLAN-DR' summary format to isolate agent deliberation.
Capability inventory: Triggers downstream execution via the $ralph and $team skills based on the finalized plan.
Sanitization: Does not explicitly define sanitization or escaping mechanisms for the interpolated content within the prompt blocks.

ralplan