agentic-phase-workflowlisted

# Agentic Phase Workflow The shape of AI-directed development work. This skill sequences the loop that turns a ratified contract and owning Issue into a reviewed, verified delivery without losing project coherence. The loop below is the **end state of ten documented revisions** — a new project starts here rather than re-deriving it. How each revision emerged from evidence belongs to `agentic-workflow-evolution`; this skill hands you the result. Worked examples and provenance: `examples.md`. --- ## 1. Start with a spec and a phased plan (before any code) If the problem, workflow, scope, experience, or visual direction is still materially unsettled, run `agentic-product-discovery` first. The spec should record ratified product choices, not conceal unresolved discovery inside an implementation plan. Write a spec first. It costs little and becomes the canonical scope reference for the whole project — reviewers cite it, defer/fix decisions turn on it, and "in scope vs. later" stops being an argument. On the reference project it took about 30 minutes to write, and reviewers then cited its sections directly when justifying findings (`examples.md` §2). A spec has four load-bearing parts. Generic skeleton: ``` # <Project> — Spec & Phase Outline ## 1. Overview — vision, target users, success criteria (one paragraph each) ## 2. Stack — table: layer | choice | rationale ## 3. Functional Reqs — FR-1, FR-2, ... each a testable "the tool must ..." ## 4. Non