QUETREX Dashboard
ENFORCEMENT-FIRST · AI DEVELOPMENT

The gate stays shut until the code is actually green.

Quetrex is an enforcement-first development system for Claude Code. Every guarantee is a hook reading an artifact on disk — a real exit code, a real review verdict, a real security report. Not a sentence from a model that says it's fine.

Open the dashboard → Install the plugin runs on your own Claude
01 THE FOUR GUARANTEES

Four promises. None of them made of trust.

Everything below is enforced by a deterministic hook that reads an on-disk artifact. If the artifact isn't there or the exit code is wrong, the pipeline does not advance. There is no "the agent said it was fine."

G1ENFORCED

No broken code ships

The verify-gate blocks "done" while typecheck, lint, build, or tests are red. Real exit codes, pinned to the commit under review. Green is proven, never claimed.

hook verify-gate · reads exit codes @commit · any ≠ 0 → blocked
G2ENFORCED

Merge is auto-gated

A separate reviewer runs in fresh context — no memory of writing the code. A clean review auto-merges. Issues go back as rework. Risk escalates to a human. Only production deploy stays manual.

hook reviewer · fresh-context verdict → auto-merge | rework | escalate
G3MANDATORY

Security is forced by detection

Touch a sensitive path — auth, secrets, migrations, payment — and a security review is triggered automatically. You can't opt out of it. A Critical finding hard-blocks the merge. No override.

hook path-detector → security-review · Critical = hard block
G4ROUTED

Simple tasks stay fast

A right-size router keeps trivial work off the full pipeline. A one-line copy fix doesn't wait behind an architect and a reviewer. Rigor where it counts; speed where it doesn't.

hook right-size-router · low-risk → skip full pipeline
02 HOW IT WORKS

Every arrow is a gate. Every gate reads an artifact.

Work moves through six stages. Between each pair sits a gate that reads a file the previous stage wrote — a verify log, a review verdict, a security report. The pipeline can't advance on a summary or a promise. It advances on the artifact, or it stops.

S1architect
Turns the task into an implementation plan and a strict file-ownership map, so parallel developers never touch the same file.
writes plan.mdownership.map
S2developer(s)
One or more developers build in parallel worktrees on separate sub-branches, each owning distinct files. They run verify locally as they go.
writes diff+verify.json
S3QA
Proves green with actual exit codes — it re-runs the checks itself and never takes a developer's word. Red here means back to S2.
reads verify.jsonexit 0 required
S4reviewer
Reads the full diff in fresh context for logic errors, architecture violations, and correctness. Emits a verdict the merge gate obeys.
writes review.verdictmerge | rework
S5security-reviewer
Runs automatically when a sensitive path was touched. A Critical finding is a hard stop — the merge cannot proceed until it's resolved.
writes security.reportCritical = block
S6git-workflow
Opens a squash PR to main and, on a clean set of verdicts, auto-merges it. Production deploy is the one step that stays in your hands.
reads all verdictssquash PR → merge
The artifact is the source of truth. Agents can narrate whatever they like; the gates only read files on disk. That's the whole design — determinism where it matters, so a good story can't ship bad code.
03 THE PLUGIN FAMILY

One engine. Every stack you actually ship.

Quetrex installs into Claude Code as a plugin family: a stack-agnostic engine that owns the pipeline and the gates, plus a language pack that teaches it your toolchain's verify commands, conventions, and sensitive paths.

ENGINE · STACK-AGNOSTIC

quetrex-factory

The pipeline, the hooks, the gates, the router. It knows nothing about your language and everything about enforcement. Drop in a pack and it inherits the toolchain.

Next.jstsc · eslint · next build · vitest
Pythonmypy · ruff · pytest
Rustcargo check · clippy · test
Swiftswiftc · swiftlint · xctest
04 BRING YOUR OWN COMPUTE

Your subscription runs it. Quetrex just refuses to let it fail.

Quetrex is the framework, not the model. Every run executes under each developer's own Claude subscription — no shared inference bill, no rate-limit contention, no vendor lock on the intelligence layer. You own the compute; the system owns the discipline.

  • InstallAdd the plugin, pick your pack. One command in Claude Code and the gates are live in the repo.
  • AnywhereTerminal-driven, run it anywhere. From your laptop or an always-on workstation over SSH — every action is a command, and the enforcement travels with the repo, not the desk.
  • CoordinateA shared kanban, not a shared babysitter. Teams watch the board and step in only at the gates — approve a merge, clear a deploy. No terminal-staring.
05 SHIP WITH CERTAINTY

Stop hoping your AI got it right.

Let the gates decide. Install Quetrex, point it at a task, and walk away — the code that reaches main passed typecheck, lint, build, tests, a fresh review, and, when it mattered, security. Because a hook checked, not because a model promised.