circuit

circuit-next

Status: Tier 0 scaffold complete; Phase 1 contract authorship closed at Slice 27 (workflow.md v0.2); Phase 1.5 Alpha Proof closed at Slice 31a (ceremony commit, 2026-04-21) per ADR-0001 Addendum B as amended by ADR-0006; Phase 2 — Implementation closed at Slice 101 (2026-04-24) for the first working workflow product spine. Full first-generation Circuit parity remains future work. CC#14 closure is a one-time waiver + retarget per ADR-0006: (14a) operator product-direction check at specs/reviews/phase-1.5-operator-product-check.md, (14b) existing Claude + Codex LLM stand-in sections in specs/reviews/phase-1-close-reform-human.md with F17 weaker-evidence flag. The canonical non-LLM human cold-read is not satisfied; weaker evidence of different shape is substituted. ADR-0006 §Precedent firewall governs any future retarget.

Claude Code plugin that automates common developer and creative workflows through a configurable, evidence-based methodology.

Why this exists

The first-generation Circuit evolved organically and accreted complexity. circuit-next is a from-scratch rewrite driven by a deliberate methodology — Contract-First + Tiny-Step-Ratcheting + Architecture-First + narrow cross-model challenger — that was itself designed through a tournament-rigor Explore run.

The operator-facing value proposition is unchanged: a small number of high-leverage workflow shapes (Explore, Build, Repair, Migrate, Sweep) with per-step configurability of model, reasoning effort, skills applied, and user-global defaults/overrides.

Methodology

See specs/methodology/decision.md for the tournament decision that defines the four methodological pillars, their stress-test survival, accepted risks, and reopen conditions.

Current phase

Phase 2 — Implementation (closed for the first working workflow, 2026-04-24 at Slice 101). Phase 0 Evidence Loop closed; Phase 1 contract authorship closed at Slice 27; Phase 1.5 Alpha Proof closed at Slice 31a. Phase 2 proved the product spine: Claude Code plugin commands reach the runtime, Explore runs through the canonical workflow, Review is wired as a second workflow shape, Claude and Codex subprocess dispatch both have current live smoke fingerprints, strict JSON artifacts are the accepted successor shape for step state, and config can select model and effort. This does not claim full first-generation Circuit parity. Build, repair, migrate, sweep, custom workflow authoring, and a polished workflow-configuration experience remain future work. specs/parity-map.md records the current gap map and recommends opening Build next; specs/plans/build-workflow-parity.md is the operator-signed Build plan now being implemented. CC#14 was amended by ADR-0006 as a one-time waiver + retarget (see Status above); the canonical non-LLM cold-read is not satisfied, and weaker evidence of different shape is substituted — carried openly on every authority surface. CC#15 preservation rests on (i) the 14a operator product-direction check and (ii) CC#13 closure by the Slice 29 property fuzzer at tests/properties/visible/, a structurally-different non-LLM mechanical probe.

See PROJECT_STATE.md for the live snapshot of where the project is. Per-slice narrative history is in PROJECT_STATE-chronicle.md (split from PROJECT_STATE.md at Slice 67; the chronicle is non-authoritative history, live state lives at ## §0 of PROJECT_STATE.md).

Layout