The Agent Skills Directory

[SAFE]: The skill coordinates a collaborative planning session among three AI roles. The logic focuses on reaching consensus through sequential review and critique, which is a defensive design pattern for high-quality output.
[COMMAND_EXECUTION]: The skill uses the delegate_task tool to orchestrate sub-agent roles. This is a standard feature of the OMH framework and is used here to facilitate planning and review rounds rather than arbitrary command execution on the host system.
[DATA_EXPOSURE]: Persistent state and consensus plans are stored locally in the .omh/ directory. No network exfiltration or access to sensitive credentials (like SSH or AWS keys) was found.
[PROMPT_INJECTION]: The skill facilitates an indirect injection surface by ingesting codebase files to provide context for sub-agents.
Ingestion points: Phase 0 (Context Gathering) reads project files into a summary.
Boundary markers: None explicitly defined in the task delegation templates.
Capability inventory: Uses delegate_task and the terminal toolset as defined in the frontmatter.
Sanitization: No input sanitization or explicit delimiters are used for the project context. This surface is intrinsic to the skill's purpose as a project-aware planner and does not present an immediate risk as the output is a document for user review.

omh-ralplan