What problem does it solve? When an AI agent starts coding immediately from a loose request, it silently guesses unstated requirements, produces code nobody can review against a written target, and lets project docs drift from reality. This orchestrator turns a vague idea into a reviewable plan with a human approval gate before any code is written, and always knows which phase a cycle is in by reading the filesystem instead of conversation memory. ## Core Features & Use Cases - Phase detection from files: Inspects cycles/ artifacts (request.md, plan.md, tasks.md, verification.md, validation.md) through an ordered rule table to determine whether the cycle is at Start, Capability, Refine, Gate, Implement, Verify, Validate, Promote, or Done. - Routing to phase skills: Hands off to the seven sibling skills (sdd-start, sdd-capability, sdd-refine, sdd-implement, sdd-verify, sdd-validate, sdd-promote) that each produce their own artifacts. - Broken-state handling: Detects contradictory states (draft plan with checked tasks, missing status: field, stale validation passes) and stops to ask the human instead of guessing. - Use Case: You return to a project after two weeks and ask "onde parei?" — the skill reads the cycle folder, finds plan.md approved with unchecked tasks, and routes directly to implementation. ## Quick Start Ask the assistant to start a new feature cycle by describing what you want to build, or ask "where did I stop?" in a repository containing a cycles directory.
