skills/flywheel/SKILL.md
Outer-loop shipping orchestrator. Composes /shape, /implement, /yeet, /deliver --polish-only, /ship, and /monitor per backlog item. Closure (archive, reflect, harness routing) lives in /ship; flywheel does not invoke /reflect directly. Use when: "flywheel", "run the outer loop", "next N items", "overnight queue", "cycle". Trigger: /flywheel.
npx skillsauth add phrazzld/spellbook flywheelInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Compose cycles of: pick a backlog item → /shape (if unshaped) →
/implement → /yeet → /deliver --polish-only → /ship → /monitor →
loop.
Abbreviated form using the convenience composer:
pick → /deliver → /yeet → /ship → /monitor → loop.
(/deliver = /shape → /implement → clean loop; /deliver --polish-only
runs that same clean loop on an existing branch.)
Migration: the polish step was
/settleuntil backlog 080 collapsed it into/deliver --polish-only;/settleis a deprecated redirect for one release.
You already know how to do each of these. This skill exists only to encode the invariants that aren't inferable from the leaf names.
backlog.d/. Flywheel has none..harness-kit/flywheel.yaml exists, load it once at cycle start with the
Harness Kit config loader and use it only for cycle tuning: cadence,
max_cycles, token budget, backlog include filters, and stop conditions.
If absent, use invocation flags and built-in defaults..harness-kit/agents.yaml, /flywheel expects receipts for whatever
lanes the leaf skills judged necessary, but it does not dispatch
providers directly./ship,
plus any provider-health follow-ups surfaced by failed or low-signal lanes./ship owns closure: squash-merge, backlog archive, /reflect, and
applying reflect's outputs. Flywheel does not invoke /reflect directly.git status --short --branch --untracked-files=all shows paths or the
shipped branch is unpushed/diverged from its remote. Verify remote sync with
git rev-list --left-right --count <branch>...<upstream> or the branch's
equivalent upstream check. Nonzero entries are action items for /yeet,
/ship, a move-out, durable ignore, or an explicit blocker. See
harnesses/shared/AGENTS.md (Closeout)./ship routes them to
harness/reflect-outputs for human review.delegate on judgment per the shared Roster contract: native subagents
by default; add cross-model critics, roster providers, or sprite lanes
(/sprites) only when they answer a distinct question. See
harnesses/shared/AGENTS.md (Roster).
Local lane guidance: Leaf skills own dispatch; /flywheel verifies phase receipts and uses lanes only for cycle strategy, risk critique, and closure-state review.
/deliver's receipt is the contract — don't peer inside.backlog.d/ but already shipped in git. Fix
the stale entry before starting a cycle on it./flywheel runs in the same worktree collide on git state. Use
separate worktrees for parallelism./reflect invocation — that's /ship's job.Semantic waiver: /flywheel composes other phase receipts rather than owning a
standalone deterministic transform. Verify a cycle by the /deliver, /ship,
/deploy, /monitor, and /reflect receipts it links, plus clean-tree
closeout and remote-sync evidence.
testing
Capture one compounding repo-technical learning while a solved problem is still fresh. Use when: after a bug fix, diagnosis, delivery, review, or incident reveals a reusable pattern worth adding to `docs/solutions/`. Trigger: /compound, /capture-learning, /learning.
testing
Route Misty Step factory application capabilities. Use when choosing, auditing, integrating, or operating Canary, Powder, Landmark, Aesthetic, or Bitterblossom: production observability, incidents, health checks, error logging, backlog/work-card state, release intelligence, UI/UX system adoption, or supervised/unsupervised agent dispatch. Trigger: /factory-apps, /factory-stack.
testing
Prove a skill beats no-skill with a falsifiable A/B eval, or retire it. Design, generate, run, and maintain a skill-specific eval: name the one claim the skill must earn, run it skill-on vs raw same-model, grade blind with objective checks first, return a keep/adapt/cut verdict. Use when: "eval this skill", "does this skill help", "prove the skill beats no skill", "write an eval for", "benchmark a skill", "is this skill worth it", "skill A/B", "skill regression test", "generate skill evals". Trigger: /skill-eval, /eval-skill, /prove-skill.
tools
> Template. Copy to `<target-repo>/.agents/skills/<repo>-<domain>/SKILL.md` > and fill every bracketed placeholder from the live target repo. Delete this > line and every other `> ` guidance line before committing. See > `../../references/repo-local-skill-generation.md` for the full process. --- name: <repo>-<domain> description: | [One paragraph: what this skill verifies/runs/operates for <repo>, stated in terms of the repo's real shape (service/CLI/library/etc.), not generic process. En