Toque takes a change from idea to release with a gate at every step. A six-stage planning workflow for Claude Code, built around an adversarial design gate that has to be passed rather than scored. Stack-agnostic. Works on any codebase.
| Plugin | What it does | Who installs it |
|---|---|---|
toque |
Six-stage AI-Native SDLC planning (intent, spec, plan, test, release, maintain) with an adversarial verifier-first design gate, plan-linked troubleshooting, documentation generation | Developers living in docs/plans/ daily |
Upgrading from 10.x? The codebase-audit and AI-readiness commands are no
longer part of Toque. Nothing replaces them here. Toque still reads analysis
files left in docs/audit/ by whatever produced them, and works fine when there
are none — see Optional inputs.
Toque asks three questions about your codebase:
| Pillar | Question | Time Orientation |
|---|---|---|
| 1. Documentation as the Foundation | What do we have? | Past |
| 2. Phased Delivery Over Big-Bang Releases | What shape is it in? | Present |
| 3. Operational Readiness | Can we safely change it? | Future |
The methodology behind all three plugins lives in METHODOLOGY.md.
The largest feature, and the reason most people install the toolkit.
/toque:plan runs six stages from idea to release. Each stage reads the
previous artifact and commits its own — the chain of artifacts is the audit
trail: who asked, what was produced, who approved.
| # | Stage | Commits | Gate |
|---|---|---|---|
| 1 | Plan | intent.md |
A named product owner accepts |
| 2 | Design | spec.md, audit.md, evidence/ |
Scope lock, then the design gate |
| 3 | Build | plan.md, code, impact-review.md |
plan.md approved before code |
| 4 | Test | test-plan.md, results |
Automated tier passes, manual tier human-confirmed |
| 5 | Deploy | review.md |
A named human authorizes. The agent never deploys |
| 6 | Maintain | A new intent.md from incidents |
None — this stage never completes |
Four rules hold in every stage: human gates are real gates, nothing is
implemented without an approved plan.md, the agent verifies its own work
before asking for review, and the agent never crosses the production gate.
Stage 2 ends in a gate that can refuse. A fresh isolated auditor reviews the spec; a canary plants a known defect to prove the audit actually looked, and every citation is re-read from disk and compared byte-for-byte.
PASS = CANARY_OK AND EVIDENCE_OK AND VERIFIED AND INFRA_OK
No weighted sum, no partial credit.
→ The plan workflow — all six stages in depth → The design gate — including its known limitation → The plan workspace — the folder, resume, staleness
Prerequisite: Claude Code installed (claude.ai)
Step 1: Open a terminal (not inside Claude Code) and add the marketplace:
claude plugin marketplace add krwhynot/toqueStep 2: Install it:
claude plugin install toque@toque-marketplace --scope userStep 3: Start Claude Code in any project and verify:
/toque:help
/toque:help maps every command.
Upgrading from a pre-10.0.0 install, or want every install path? → Install
/toque:quick-planon something small, to see the shape of the output/toque:planwhen the work is big enough to want the gate/toque:plan-statusto pick up where you left off
If some other tool has left a risk assessment, dependency map, feature
inventory or integration scan in docs/audit/, Toque reads them, and a plan
grounded in those beats one written from the code alone. It never requires
them.
New to it? Quickstart walks the first plan end to end, and suggests two lighter commands to try first.
| Page | What's in it |
|---|---|
| Quickstart | Your first plan, end to end, with what appears on disk |
| Install | Every install path, scopes, upgrading, what breaks without Node |
| When to use Toque | Use / don't use per plugin — including where it is overkill |
| The plan workflow | All six stages in depth, approval tiers, parallelism |
| The plan workspace | Plan folder anatomy, status.json, resume, staleness cascade |
| The design gate | Canary, evidence validation, the pass expression, the limitation |
Deeper references: GUIDE.md · METHODOLOGY.md (the theory) · interop.md (optional inputs Toque reads but does not write) · CHANGELOG.md
.claude-plugin/marketplace.json # the catalog entry and its ref+SHA pin
plugins/toque/ # planning core (commands, agents, skills, hooks)
documentation/ # task-shaped user documentation
tests/ # the suite (run-all.sh, seven layers)
docs/ # dev-time records: plans, specs, release notes
METHODOLOGY.md # the methodology reference (not shipped by the plugin)
interop.md # optional docs/audit/ inputs Toque reads
Every plugin manifest carries the same version, bumped together by
.github/release.sh — the catalog entry always pins one ref and one SHA.
Toque runs its hooks and design-gate tools as Node scripts and requires Node.js 18 or later — the same runtime Claude Code itself needs, so if Claude Code runs, it does too.
The always-on safety hooks that used to ship as toque-guard (force-push
and DB-deploy blocking, migration protection, change/test tracking) were retired
in 9.0.0. Claude Code's own permission rules cover the same commands with no
runtime dependency; see METHODOLOGY.md §6
for the recommended settings.json rules.
See CHANGELOG.md for full history.
Current: v11.0.0
See CONTRIBUTING.md for guidelines.
MIT. See LICENSE.