BuildProven · Open source

Claude Kit

A Claude Code + OpenAI Codex toolkit whose quality gate reads your diff, scores its risk, and scales its own review depth to match.

Install free View source on GitHub

Quality gate · live routing

docs/README.md LOW → focused review → ship

scripts/payment-handler.js HIGH → broad + adversarial review

.github/workflows/quality.yml CRITICAL → maximum effort → merge blocked until green

Why it exists

“Please be thorough” is not a quality system.

Claude Kit turns review depth, failure handling, and merge evidence into executable policy. Small changes stay fast. Risky changes earn scrutiny.

Risk decides effort Deterministic scoring reads the actual diff. Security, workflow, and logic changes escalate even when the patch looks small.
Loops stay bounded A run governor caps review rounds, elapsed time, and fix commits. Exhausted or inconclusive reviewers fail visibly instead of looping forever.
Evidence follows the merge Revision-bound review trailers, required CI, and unresolved-finding gates prevent “review ran” from being mistaken for “review passed.”

What you get

The complete setup. No paid layer.

The public repository includes the commands, skills, agents, hooks, and automation. Nothing essential is held back.

Risk-adaptive quality Review panels, adversarial verification, merge gates, and quota-aware Claude/Codex fallback.
Autonomous workflow Bounded backlog execution, investigation, planning, testing, and recovery loops.
Specialist judgment Security, architecture, performance, strategy, legal, SEO, and content review capabilities.
Repository guardrails Destructive-path protection, branch controls, defensive linting, and visible failure handling.
Native agent integration A namespaced Claude Code plugin plus native Codex skills, shared instructions, worktree isolation, and provider-aware orchestration.
Tested infrastructure More than 430 automated tests plus formatting, security, license, pattern, and skill-size gates.

Choose your coding agent

Install for Claude Code or Codex.

Claude Code

Install the namespaced /bs:* plugin.

/plugin marketplace add buildproven/claude-kit
/plugin install bs@buildproven

OpenAI Codex

Install the curated skills in Codex's native Agent Skills format.

git clone https://github.com/buildproven/claude-kit.git
cd claude-kit
bash scripts/setup-codex-skills.sh

Keep your existing ~/.codex/AGENTS.md. A provider-neutral instruction template and dedicated Codex-only installer are planned. Free and open source: most of the kit is MIT; one substantially modified upstream skill retains its Apache-2.0 license.

Built in the open

Use the system. Inspect every rule.

Pick your coding agent, read the implementation, and keep only the guardrails that make your work better.

Install Claude Kit Read the v4.0.0 release