What problem does it solve?
Provides a structured root brain for project job orchestration that decides which job to work on, ensures each job has the necessary context, and enforces job hygiene so work proceeds with clarity and evidence.
Core Features & Use Cases
- State classification: Detects IDLE, HEALTHY, STALE, OVERLOADED, and GROWTH_NEEDED states from project metrics.
- Phase orchestration: Runs a 12-phase Markov-style loop (session recovery, context load, priority assess, focus decide, delegate, monitor, evidence collect, job discovery/create, self-improve).
- Context management: Reads and enriches .claude/CLAUDE.md and .claude/data.json, updates focused_jobs, and logs transitions to a transitions file for learning.
- Scripted tooling: Integrates shell utilities (state-classifier.sh, job-metrics.sh, transition-tracker.sh, decide-next-phase.sh) to produce human- or machine-readable recommendations and metrics.
- Use Case: Use this Skill to automatically pick the next engineering task from a backlog, prepare a condensed working context for the programmer skill, and track evidence and outcomes across sessions.
Quick Start
Load the project's .claude/data.json and .claude/CLAUDE.md, run decide-next-phase.sh, and apply the recommended focus change to update focused_jobs.