The Agent Skills Directory

[COMMAND_EXECUTION]: The skill manages implementation tasks through CLI tools including omc team and the well-known cursor-agent. It also utilizes the Linux flock utility for kernel-level advisory locking of state files.
[DYNAMIC_EXECUTION]: As its primary intended function, the skill autonomously generates, builds, and tests code based on user-provided ideas. This execution is scoped within a structured multi-phase lifecycle.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user ideas to drive the development process. It implements robust mitigations through a mandatory multi-perspective validation phase (Phase 4), which includes a dedicated security-reviewer subagent to verify the generated code for vulnerabilities.
Ingestion points: User-supplied product ideas or task descriptions provided as arguments to the skill.
Boundary markers: Utilizes session-scoped state and transcript evidence boundaries to isolate execution contexts.
Capability inventory: Spawns specialized subagents (architect, security-reviewer, code-reviewer), writes files to the .omc/ directory, and executes code via the team runtime.
Sanitization: Employs mandatory functional, security, and quality reviews that must all reach consensus before completion.
[SAFE]: The skill accesses ~/.config/claude-omc/config.jsonc solely to retrieve its own configuration settings and project-specific overrides, following standard user-level application practices.

autopilot