Composable process framework for LLM-driven work at scale: durable process record, mechanical capture & priming, honest telemetry. MCP plugin for Claude Code and compatible agents.
Skill Claude CodeCodex
Autonomous SDLC loop that runs execute → review → refine over the work board until the project converges — zero open claimable work and no unresolved critical/significant findings — or a cycle/appetite limit is hit. Unattended: Andon escalations go to a proxy-human agent instead of stopping for the user.
Skill Claude CodeCodex
Execute the work board produced by init/refine. Claims claimable items under lease-fenced discipline, delegates each to a worker subagent, runs incremental review, handles findings by severity, and flags unresolvable issues to the user via an Andon stop. Builds work; does not design or re-plan.
Skill Claude CodeCodex
Set up an ideate project. Creates the workspace, enables the steering store, captures the project's intent and guiding principles, and — for an existing codebase — surveys the current architecture. It lays the foundation; it does NOT create work items. Decomposing ideas into work is refine's job.
Skill Claude CodeCodex
Decompose an idea into actionable work. The first working step after init and the recurring planning engine: takes a described goal, change, or set of review findings, analyzes it against the existing code and steering, and produces board work items with dependency and containment edges. Plans work; execute builds it.
Skill Claude CodeCodex
Comprehensive review of completed work. Spawns specialized reviewers in parallel (code, spec, gap), synthesizes and records their findings by severity, and curates steering. Supports cycle review (default), domain review (--domain name), full audit (--full), and ad-hoc scope (natural language).