The Agent Skills Directory

[SAFE]: The skill implements a robust requirements-gathering phase that acts as a safety gate, preventing the agent from performing potentially incorrect autonomous actions based on vague prompts.\n- [DATA_EXPOSURE]: The skill reads configuration files from standard user and project paths (~/.claude/settings.json, ~/.config/claude-omc/config.jsonc) to resolve the ambiguity threshold. This is standard behavior for configuring tool-specific parameters.\n- [COMMAND_EXECUTION]: The skill orchestrates other agent skills (such as plan, autopilot, and ralph) using the platform's Skill() invocation. These bridges are explicitly gated by user approval steps and are not triggered automatically.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests user ideas and codebase context. It mitigates potential injection by summarizing untrusted input before processing and strictly defining the boundaries of the generated specification.

deep-interview