← ClaudeAtlas

epic-flywheellisted

Drive an entire epic to completion semi-autonomously — runs the per-story flywheel (create → dev → review) for every story with granular commits, then a single epic-boundary verification + manual-test-planning pass. Use when the user says "epic flywheel", "run the whole epic", "flywheel the epic", or "run epic {N} autonomously".
rterakedis/leanwheel-skills · ★ 2 · AI & Automation · score 65
Install: claude install-skill rterakedis/leanwheel-skills
# Epic Flywheel Skill **Goal:** Take a whole epic from "stories not started" to "all stories implemented, reviewed, and verified together" with minimal steering. The per-story automated gates (Build & Test, evals RUN, invariant verification) still stop the loop on a real compounding bug; the **manual / integration test pass is deferred to the epic boundary** (DD-30). **Relationship to `/story-flywheel`:** the autonomous, epic-scoped layer above it — same three subagents and model routing (story-flywheel → **Subagent Delegation & Model Routing**; not duplicated here). Adds: 1. commit-per-step so a bad story can be unraveled; 2. within-epic auto-advance on clean stories; 3. an **Epic Boundary Gate** (build+test+evals+invariants across the whole epic, **HALTs for help** on any failure); 4. continuous deferred-item re-homing; 5. a deduplicated **rolled-up test plan** that subtracts what automation already proves, then splits the rest into simulator-runnable vs physical-device-required. **Phase files** — read each only when its phase is reached, and re-read it then rather than relying on memory (a long epic compacts this context): - Epic boundary, gates 1–4c and the step 6 report → [boundary-gate.md](boundary-gate.md) - Epic boundary, step 5 test plan → [test-plan-rollup.md](test-plan-rollup.md) - Epic boundary, step 5b, only when its condition holds → [squash-merge.md](squash-merge.md) --- ## Iron rules 1. **HALT, don't push forward.** A red Build & Test Gate, a failing eva