bkit — AI Native Development OS

The only Claude Code plugin that verifies AI-generated code against its own design specs.

What bkit gives you

bkit turns AI-driven coding into a verified PDCA workflow. You describe intent, bkit drives Plan → Design → Do → Check → Act, and gap-detector closes the loop by comparing the design back against the implementation.

Verified, not just generated:

• PDCA as a real state machine — 20 transitions, L0-L4 automation, M1-M10 quality gates
• Match Rate ≥ 90% between design specs and shipped code, enforced by gap-detector; sub-90 triggers pdca-iterator auto-improvement (max 5 cycles)
• CTO-Led Agent Teams orchestrate pm-lead / qa-lead / pdca-iterator for non-trivial work
• Defense-in-Depth 4-layer — CC sandbox → bkit PreToolUse hooks → audit-logger sanitizer (OWASP A03/A08) → Token Ledger NDJSON
• Docs=Code CI keeps README · CHANGELOG · plugin manifest · hooks · session intro in lockstep, plus a 5-location One-Liner SSoT and a 5-location BKIT_VERSION invariant
• Invocation Contract L1~L5 — 226 assertions guarding skill/agent/hook surface across releases

Quick start

# 1. Install via Claude Code marketplace
claude plugin install bkit

# 2. (Optional) Enable Agent Teams for the cto-lead orchestration
export CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1

# 3. In Claude Code, drive a feature end-to-end
/pdca pm my-feature       # PM Discovery → PRD
/pdca plan my-feature     # PDCA Plan
/pdca design my-feature   # PDCA Design
/pdca do my-feature       # Implementation
/pdca check my-feature    # gap-detector → Match Rate
/pdca act my-feature      # Completion report

First-time users see an interactive 3-minute tutorial on the very first session. Recommended runtime: Claude Code v2.1.118+ (79 consecutive compatible releases since v2.1.34); minimum v2.1.78.

Architecture (v2.1.12)

Clean Architecture 4-Layer (Domain ports/guards/rules · Application · Infrastructure · Presentation) with 7 Port↔Adapter pairs (cc-payload · state-store · regression-registry · audit-sink · token-meter · docs-code-index · mcp-tool) and a 3-Layer Orchestration core (intent-router · next-action-engine · team-protocol · workflow-state-machine). v2.1.11 added 4 Sprints (α Onboarding · β Discoverability · γ Trust · δ Port + Governance) covering 20 FRs; v2.1.12 hotfix patches the lib/evals/runner-wrapper.js argv mismatch (silent failure of /bkit-evals run) plus false-positive defense and JSON-parse robustness.

Documentation

• README-FULL.md — full feature list, complete version history, deep architecture
• CHANGELOG.md — release history
• docs/01-plan/ — PDCA Plan documents (Korean)
• docs/02-design/ — design specs + ADRs
• docs/03-analysis/ — analysis evidence
• docs/04-report/ — completion reports
• docs/05-qa/ — QA outputs

License

Apache 2.0 — see LICENSE.

POPUP STUDIO PTE. LTD. · [email protected]