Contents

• Why KyZN?
• Quick Demo
• Quick Start
• Usage
• How It Works
• Safety Model
• Project Structure
• Contributing
• License

---

Why KyZN?

Improving a codebase with Claude is powerful — but doing it manually means you're the glue holding the workflow together:

1. Run linters, type checkers, and security audits for your language
2. Read the output, decide what matters
3. Paste findings into Claude with enough context
4. Hope Claude doesn't burn tokens on cosmetic renames and import shuffling
5. Review the diff for regressions and leaked secrets
6. Run tests yourself
7. Check the health score didn't drop
8. Create a PR with a summary of what changed

KyZN replaces all of that with one command. It runs real tools, scores your repo, dispatches 4 specialist agents to find issues across security, correctness, performance, and architecture — then fixes them in severity batches with build verification after each one. If something breaks, it auto-retries. If the health score drops, it aborts. When it's done, you get a PR with before/after scores.

kyzn fix   →  profile repo  →  4 Opus specialists  →  consensus  →  Sonnet fixes  →  verify  →  PR

Supports Node.js, Python, Rust, and Go out of the box. Works on any project type for generic analysis.

How KyZN uses tokens efficiently

KyZN is designed to get more value per token than an interactive Claude session:

Mechanism	What it does
Structured JSON input	Linter output is parsed into scored JSON — Claude gets signal, not 200 lines of raw tool output
Mode constraints	deep mode blocks cosmetic changes (renames, reformats, import reordering) so tokens go to real fixes
Read-only analysis	Specialist agents only get Read/Glob/Grep — zero tokens spent on exploratory edits during analysis
Cached profiler	Repo conventions are profiled once per commit SHA and reused across runs
Consensus dedup	4 specialists may flag the same issue — consensus removes duplicates before the fix phase starts
Hard budget caps	Every Claude invocation has --max-budget-usd and --max-turns enforced (default quick run: $2.50, 30 turns)
Stateless sessions	--no-session-persistence on every call — no cross-run context bloat accumulating
Structured fix plans	Each finding includes target file, function, and pattern — the fix agent doesn't spend tokens figuring out where to edit

---

Quick Demo

$ kyzn measure

  Project Health Score: 68 / 100

  security        ████████████████░░░░  80%
  testing         ██████████░░░░░░░░░░  50%
  quality         ██████████████░░░░░░  72%
  performance     ████████████████████ 100%
  documentation   ████████████░░░░░░░░  60%

$ kyzn fix

  → Profiler: scanning repo conventions...
  → 4 specialists dispatched (security | correctness | performance | architecture)
  → Consensus: 27 findings (deduped from 32)
  → Fixing HIGH (7 issues)... ✓ Build passes
  → Fixing MEDIUM (10 issues)... ✓ Build passes
  → Fixing LOW (6 issues)... ✓ Build passes
  → PR created: https://github.com/you/project/pull/5

One command. Zero config. Real bugs fixed, verified, and shipped.

---

Quick Start

Prerequisites

Tool	Required	Purpose
git	Yes	Branch management
gh	Yes	PR creation (GitHub CLI)
claude	Yes	AI analysis (Claude Code)
jq	Yes	JSON processing (auto-installed with checksum verification)
yq	Yes	YAML config (auto-installed with checksum verification)

macOS: Requires Bash 4.3+ (brew install bash). The system /bin/bash is v3.2 and will not work.

Install

curl -fsSL https://raw.githubusercontent.com/bokiko/KyZN/main/install.sh | bash

Or clone manually:

git clone https://github.com/bokiko/KyZN.git ~/.kyzn-cli
ln -sf ~/.kyzn-cli/kyzn ~/.local/bin/kyzn

First Run

kyzn doctor     # Check prerequisites
kyzn init       # Interactive setup → .kyzn/config.yaml
kyzn measure    # See your health score
kyzn fix        # Deep analysis + auto-fix → PR

---

Usage

kyzn fix — The main command

kyzn fix                        # Full pipeline: profile → analyze → fix → verify → PR
kyzn fix --auto                 # Non-interactive (cron-safe)
kyzn fix --profile hybrid       # Opus for security+correctness, Sonnet for perf+arch
kyzn fix --min-severity HIGH    # Only fix HIGH+ findings
kyzn fix --fix-budget 10.00     # Budget for fix phase

Profiler scans conventions, 4 Opus specialists find issues in parallel, consensus deduplicates, Sonnet fixes in severity batches (CRITICAL → HIGH → MEDIUM → LOW) with build/test verification after each batch. If a fix breaks the build, reflexion retry gives Sonnet a second chance with the error output. Opens a PR when done.

kyzn analyze — Report only (no changes)

kyzn analyze                    # 4 Opus specialists + consensus report
kyzn analyze --focus security   # Single specialist (security only)
kyzn analyze --single           # Single general reviewer (cheaper)
kyzn analyze --export report.md # Export to custom path

kyzn quick — Lightweight single-pass

kyzn quick                      # Single Sonnet pass, fast
kyzn quick --auto               # Non-interactive
kyzn quick --mode deep          # Real improvements only
kyzn quick --mode clean         # Dead code + naming cleanup
kyzn quick --mode full          # Everything

Other commands

kyzn measure                    # Health score only
kyzn doctor                     # Check prerequisites
kyzn history                    # Show all runs
kyzn diff <run-id>              # Show what changed
kyzn approve <run-id>           # Sign off
kyzn reject <run-id> -r "why"   # Reject with feedback
kyzn schedule daily             # Cron at 3am daily
kyzn schedule off               # Remove schedule
kyzn status                     # Health score dashboard
kyzn dashboard                  # Machine-wide activity summary
kyzn selftest                   # Run 267 quick tests
kyzn selftest --full            # Run 276 tests (incl. stress)

---

How It Works

kyzn fix
  │
  ├─ Detect project type (package.json / Cargo.toml / go.mod / etc.)
  ├─ Measure health score with real tools (eslint, ruff, clippy, go vet)
  ├─ Profile repo conventions (Sonnet reads your code patterns)
  │
  ├─ 4 Opus specialists in parallel:
  │   ├─ Security      ─┐
  │   ├─ Correctness    ├─→ Consensus (dedup + rank)
  │   ├─ Performance    │
  │   └─ Architecture  ─┘
  │
  ├─ Sonnet fixes in severity batches:
  │   ├─ CRITICAL → verify → commit
  │   ├─ HIGH     → verify → commit
  │   ├─ MEDIUM   → verify → commit
  │   └─ LOW      → verify → commit
  │   (failed batch → reflexion retry → revert if still fails)
  │
  ├─ Score regression gate
  └─ Push branch → create PR

Health score (out of 100): security 25%, testing 25%, quality 25%, performance 15%, documentation 10%. Configurable.

Languages: Node.js (eslint, tsc, vitest/jest), Python (ruff, mypy, pytest), Rust (clippy, cargo test), Go (go vet, go test). Generic works on anything.

---

Safety Model

KyZN runs AI with real tool access on your code. Every layer has safety constraints:

Layer	Protection
Branch isolation	All changes on kyzn/ branches, never touches main
Hook protection	All git operations disable hooks via core.hooksPath=/dev/null
Tool allowlist	Per-language restrictions tightened to specific subcommands (glob-safe where possible)
File restrictions	Claude cannot read ~/.ssh, ~/.aws, .env, key files, Terraform state
Symlink detection	Rejects repos with symlinks escaping the repo root
Budget cap	Hard ceiling: $25/run, 100 turns, 10000 diff lines
Build gate	PR only if build + tests pass
Score gate	Aborts if health score drops
Secret detection	Unstages files matching .env, .pem, .key, credentials patterns
CI blocking	Workflow files unstaged by default
Trust isolation	Autopilot stored in gitignored local.yaml (not poisonable via commits)
Supply chain	jq and yq verified with SHA256 checksums on install and in CI
Prompt hardening	Project names sanitized, raw data fenced to prevent prompt injection
Concurrency lock	Atomic mkdir-based lock with stale PID detection prevents concurrent runs

Important: KyZN executes your project's build and test commands. Do not run on repos you don't trust. See SECURITY.md for the full threat model.

---

Project Structure

kyzn/
├── kyzn                    # Entry point + subcommand routing
├── install.sh              # Installer (checksum-verified deps)
├── lib/                    # 13 core modules
│   ├── core.sh             # Logging, config, constants
│   ├── detect.sh           # Project type detection
│   ├── measure.sh          # Health score computation
│   ├── execute.sh          # Claude invocation + safety
│   ├── analyze.sh          # Multi-agent pipeline + fix phase
│   ├── verify.sh           # Build/test verification
│   ├── prompt.sh           # Prompt assembly
│   ├── allowlist.sh        # Per-language tool restrictions
│   ├── report.sh           # PR body generation
│   ├── interview.sh        # Interactive setup
│   ├── history.sh          # Run history + dashboard
│   ├── approve.sh          # Approve/reject workflow
│   └── schedule.sh         # Cron scheduling
├── measurers/              # Per-language health measurers
│   ├── generic.sh, node.sh, python.sh, rust.sh, go.sh
├── templates/              # System prompts + analysis templates
├── profiles/               # Focus-specific prompts
├── tests/selftest.sh       # 276 tests (quick + stress)
├── SECURITY.md             # Threat model + published audit
└── .github/workflows/      # CI (ShellCheck)

---

Contributing

KyZN is early-stage and actively developed. Contributions are welcome — whether it's a bug fix, a new language measurer, or an idea for the pipeline.

Quick dev setup

git clone https://github.com/bokiko/KyZN.git
cd KyZN
bash tests/selftest.sh          # 267 quick tests (~3s)
shellcheck -S warning kyzn lib/*.sh measurers/*.sh tests/selftest.sh

No build step — it's pure Bash. See CONTRIBUTING.md for conventions, commit format, and PR guidelines.

Where to contribute

Area	What's needed	Start here
New languages	Add measurers for Ruby, Java, PHP, etc.	measurers/ — follow python.sh as a template
Measurers	Improve scoring accuracy, add new tools	measurers/*.sh
Analysis prompts	Better specialist prompts, fewer false positives	templates/
Safety	New edge cases, threat model gaps	lib/execute.sh, SECURITY.md
Tests	Cover untested paths, new edge cases	tests/selftest.sh
Docs	Improve guides, add examples	README.md, docs/

Report a bug or request a feature

• Bug Report
• Feature Request

---

License

MIT — see LICENSE.

---

Made by @bokiko