Skip to main content

Model-agnostic quality-gate kit for AI-assisted developers

Project description

HyoDo

A model-agnostic quality-gate kit for AI-assisted developers.

HyoDo helps developers review, score, and ship AI-assisted code with a repeatable quality workflow. Primary surface is the hyodo CLI and CI gates. Optional adapters cover Claude Code, Codex, Grok, Gemini CLI, Cursor, and other agent UIs so AI-generated changes can be inspected before they become trusted code.

Model agnostic CLI and CI first Tiered routing intent only License Python Version


30-second proof

Proof point What to inspect
Public package hyodo/ Python package, pyproject.toml, and hyodo CLI entrypoint
Agent adapters commands/ slash-command docs (Claude and compatible agents) plus plain CLI
Quality gates .github/workflows/ci.yml and .github/workflows/smoke.yml
Safety posture SECURITY.md, docs/SECURITY_SURFACE.md, .env.example, .env.minimal
Routing discipline Tiered model routing is design intent only; no guaranteed savings claim

If you are reviewing HyoDo for a role, partnership, or technical screen, start with QUICK_START_SIMPLE.md, docs/PROVIDER_PROOF.md, docs/EXTERNAL_CLAIM_AUDIT.md, and the docs/ index. Claude-specific mapping is optional appendix only (docs/ANTHROPIC_PROOF.md).

Who is this for?

HyoDo is for developers and small teams who:

  • use AI-assisted coding workflows (any major model vendor or local models);
  • want a repeatable review checklist before trusting AI-generated changes;
  • need fast local checks for linting, typing, tests, and safety;
  • want to route simple tasks away from expensive model calls when possible;
  • prefer CLI/CI-first workflows that optional agent UIs can wrap.

What problem does it solve?

AI-assisted development is fast, but speed without review creates risk. HyoDo turns review into a simple operating loop:

AI-assisted change
→ hyodo check   (or /check in an agent adapter)
→ hyodo score + hyodo safe
→ fix or escalate
→ ship with a visible quality trail

The goal is not blind automation. The goal is to make AI-assisted work easier to inspect, safer to merge, and cheaper to operate.

What HyoDo does not claim

  • It does not replace human code review, tests, or security review.
  • It does not guarantee a fixed cost-reduction percentage.
  • It does not automatically approve risky changes just because a score is high.
  • It does not require the optional philosophy layer to use the practical gates.
  • It does not treat Dependabot alerts on optional/legacy afo_core as failures of the public hyodo package.

What is included?

  • CLI-first quality loophyodo check, hyodo score, hyodo safe, hyodo trinity.
  • Agent adapterscommands/ docs for slash-command UIs (Claude Code and compatible tools).
  • Quality gates — lint, format, type, test, and security-oriented checks.
  • Scoring utilities — Python package helpers for repeatable review signals.
  • Tiered model routing (intent) — optional guidance to prefer FREE/STANDARD/PREMIUM paths by risk and complexity. Not a measured savings product.
  • Public package gates — CI checks for the public hyodo package, with extended afo_core checks separated as advisory.

Quick Start

Recommended install: clone, inspect, then run

git clone https://github.com/lofibrainwav/HyoDo.git ~/.hyodo
cd ~/.hyodo

# Optional: inspect the installer before running it
sed -n '1,220p' install_interactive.sh

# Interactive setup
./install_interactive.sh

One-line install for trusted environments

curl -sSL https://raw.githubusercontent.com/lofibrainwav/HyoDo/main/install_interactive.sh | bash

Run with the CLI (recommended, model-agnostic)

cd ~/.hyodo
pip install -e ".[dev]"
hyodo start
hyodo check
hyodo score --truth 0.9 --goodness 0.9 --beauty 0.9 --benevolence 0.9 --loyalty 0.9
hyodo safe

Optional: agent slash-command adapter

If you use Claude Code or another slash-command agent that can load commands/:

cd ~/.hyodo
# open your agent of choice, then:
/start
/check
/score
/safe

Requirements

Minimal setup

  • Python 3.10+
  • Git
  • Terminal (any agent UI optional)

Full setup

  • Python 3.10+
  • Git
  • Optional agent CLI (Claude Code, Codex, Grok, Gemini CLI, Cursor, etc.)
  • Docker & Docker Compose when using extended local services
  • Redis, PostgreSQL, Ollama, or the provided Docker setup only if you need extended modules

Configuration

Minimal configuration

cp .env.minimal .env
# Set provider keys only for the adapters you actually use

Full configuration

cp .env.example .env
# Fill in the services you plan to use

Keep secrets out of git history. Never commit .env files containing real credentials.

Basic Usage

hyodo check     # Run quality gates
hyodo score     # Calculate review signal (not auto-approval)
hyodo safe      # Lightweight safety early-warning scan
hyodo trinity "describe change"  # Structured review checklist

Score Interpretation

F Score S Score Status Suggested Action
F ≥ 54 S ≥ 8 Excellent Strong review signal — human approval still required
F ≥ 45 S ≥ 7 Good Review recommended
F < 45 Needs Work Improve before merge

Scores are decision support, not a replacement for human review.

Quality Gates

4-Gate CI Protocol

Gate 1: Pyright → type checking
Gate 2: Ruff → lint + format
Gate 3: pytest → tests
Gate 4: SBOM / security-oriented seal

Tiered model routing (intent only)

HyoDo can help teams think in tiers so not every task goes to a premium model. This is routing guidance, not a cost-savings guarantee.

Tier Use Case Example providers (not exclusive) Cost Profile
FREE / local Read-only, search, lint, tests Ollama, local open models $0 where available
STANDARD Simple edits, low-risk cleanup Codex, Gemini, Grok, mid-tier APIs Low-moderate
PREMIUM Complex refactors, high-risk decisions Claude, GPT, Gemini pro tiers, etc. Higher

No public savings benchmark is linked. Do not claim fixed percentage reductions. If you need enterprise cost proof, measure your own workflows first.

Philosophy Layer: HYOGOOK V5

HyoDo also includes an optional philosophy-driven scoring model called HYOGOOK V5. It evaluates code through six review pillars and uses a geometric mean component so weak dimensions cannot be fully hidden by strong ones.

F = (T + G + In + B + C) + ⁵√(T × G × In × B × C)
S = ⁵√(T × G × In × B × C)
Pillar Weight Focus
Benevolence 25% Developer experience, user serenity
Truth 22% Technical accuracy, architecture
Goodness 18% Security, stability, ethics
Loyalty 15% SSOT compliance, project context
Beauty 15% Clean code, UX, documentation
Eternity Geometric Harmony and sustainability

Range: F ∈ [6, 60], S ∈ [1, 10]

Legacy note: WEIGHTED_V1 used an older five-pillar label set. HyoDo v3.1.0 public docs use the HYOGOOK V5 six-pillar model.

Project Structure

hyodo/              # Public Python package (release surface)
commands/           # Optional agent slash-command adapters
docs/               # Architecture, proof maps, security surface
.github/workflows/  # CI/smoke gates for public package
afo_core/           # Extended/legacy modules (advisory, not public gate)

Documentation

Document Purpose
QUICK_START_SIMPLE.md 3-minute quick start
docs/PROVIDER_PROOF.md Model-agnostic proof map
docs/SECURITY_SURFACE.md Public package vs afo_core security boundary
docs/EXTERNAL_CLAIM_AUDIT.md Measured audit of external market claims
QUICK_START.md Detailed guide
install_interactive.sh Interactive installer (English, minimal-first)
docker-compose.minimal.yml Optional extended Docker setup
SECURITY.md Security policy
CHANGELOG.md Release notes

Contributing

See CONTRIBUTING.md for guidelines.

All contributions are evaluated with practical quality gates first. The HYOGOOK V5 score can be used as an additional review signal, but passing tests, security checks, and human review remains required.

License

MIT License - see LICENSE

Links


HyoDo — safer AI-assisted development through visible quality gates.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

hyodo-3.1.5.tar.gz (22.6 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

hyodo-3.1.5-py3-none-any.whl (17.4 kB view details)

Uploaded Python 3

File details

Details for the file hyodo-3.1.5.tar.gz.

File metadata

  • Download URL: hyodo-3.1.5.tar.gz
  • Upload date:
  • Size: 22.6 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.14.6

File hashes

Hashes for hyodo-3.1.5.tar.gz
Algorithm Hash digest
SHA256 69d34ebcb02f267c0bceda2d8fe2be3b9c8ee631629fc7a22344d31b903bd608
MD5 c652e03c31ef4cc2b340ed6a7ef8ae2e
BLAKE2b-256 8317ccfebc556a7326b066b86addbf224b06bede43448369c3dca5982e44c31b

See more details on using hashes here.

File details

Details for the file hyodo-3.1.5-py3-none-any.whl.

File metadata

  • Download URL: hyodo-3.1.5-py3-none-any.whl
  • Upload date:
  • Size: 17.4 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.14.6

File hashes

Hashes for hyodo-3.1.5-py3-none-any.whl
Algorithm Hash digest
SHA256 8f1189577c775a24b8d3e614c615574004c0e69827808ab02c987bd962472306
MD5 22b5d5f14915638c1b873cd84e686475
BLAKE2b-256 6125fece9a7a547b85780c462eb09410c858876ca0b84dbe9e78993c6c02a178

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page