convez.ai
Skill
uzysjung

uzys-agent-harness

uzysjung/uzys-agent-harness

Ver no GitHub
InicialPluginScore · de 0 a 100

One-command installer & curator of vetted, Docker-verified AI-coding workflows (skills, plugins, rules, hooks) across Claude Code, Codex, OpenCode & Antigravity.

Instalação

$npx convez-ai uzysjung/uzys-agent-harness

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
3
Forks
0
Acompanhando
0
Saúde
71%
Último commit
há 3 dias
Versão
v26.95.0
Issues abertas
1
Contribuidores
1

README

A tradução para o português deste README ainda está na fila.

uzys-agent-harness

Install only the AI-coding workflows your tech stack actually needs — vetted, curated, and set up with one command across Claude Code, Codex, OpenCode & Antigravity.

Coding agents keep getting stronger out of the box — piling on skills and MCPs you'll never use just bloats their context. And the awesome-lists have too many options to wade through. agent-harness curates by tech stack: of the vetted options, you install only what this project actually calls for. Claude Code is first-class; Codex / OpenCode / Antigravity get the skills + rules layer. Project scope by default — no global pollution unless you ask.

License Version CI

agent-harness demo — one-command install of vetted AI-coding workflows

What "vetted" means — ≥ 1000 GitHub stars + active maintenance + a Docker install-verification run, re-checked by CI (catalog-verify, trust-tier-drift). It is not a line-by-line security audit or a prompt-injection scan of asset contents. npm/npx assets are version-pinned; plugin/skill assets resolve to upstream HEAD (not commit-pinned yet). Treat installed assets like any third-party dependency — see SECURITY.md.

🇰🇷 한국어


Install

npx -y @uzysjung/agent-harness

A 6-step interactive wizard guides everything. No flags needed. Safe on an existing project — it backs up your settings.json / CLAUDE.md before any change (details below); nothing is deleted.

Step 1/6  Tracks            ← pick your stack
Step 2/6  CLI               ← claude / codex / opencode / antigravity
Step 3/6  Install items     ← review pre-checked recommendations
Step 4/6  Scope             ← Project (default) or Global
Step 5/6  Confirm
Step 6/6  Installing

After install:

claude
> /uzys:spec    # only if you checked "uzys-harness 6-Gate workflow" in step 3

Non-interactive install (CI / scripts / Docker)

The wizard needs a TTY. For CI pipelines, onboarding scripts, or containers, use flags — the same path our own verification CI runs on:

npx -y @uzysjung/agent-harness install \
  --track tooling --cli claude --scope project \
  --with uzys-harness --with bmad-method
FlagMeaning
--track <name>Track to install (repeatable)
--cli <target>claude / codex / opencode / antigravity (repeatable)
--scope <s>project (default) or global
--with <asset-id> / --without <asset-id>Add / remove any catalog asset by id (repeatable) — ids in the compatibility matrix

v26.81.0 (ADR-022): per-asset flags like --with-bmad were removed — --with <asset-id> is the single opt-in surface. Behavior flags (--with-karpathy-hook, --with-codex-prompts, --with-antigravity-global, --with-prune, …) remain.

What a track actually installs (example)

Curation, not a list to browse — pick csr-supabase and step 3 pre-checks exactly these, nothing else (uncheck any before install):

TrackPre-checked assets
csr-supabasereact-best-practices · shadcn-ui · web-design-guidelines · supabase-agent-skills · postgres-best-practices · supabase-cli
datapolars · dask · python-resource-management · python-performance-optimization · anthropic-data-plugin

Full per-track matrix ↓ · or the compatibility matrix.


Installing into an existing project

agent-harness never silently overwrites your config. Before replacing an editable file whose contents differ, it writes a timestamped backup next to it — and every backup path is printed in the install summary (backup rows). Nothing is deleted.

You already have…What happens
.claude/settings.json with your own hooks / statusLineBacked up to settings.json.backup-<ts> before update
Root CLAUDE.md (yours differs from the generated one)Backed up to CLAUDE.md.backup-<ts> before the merge write
.claude/ on --reinstall / update modeThe whole directory is renamed to .claude.backup-<ts> first
.mcp.jsonYour existing MCP servers are preserved and merged, not replaced

Fresh project? None of this triggers — backups only protect pre-existing files.


Tracks

A track is a preset bundle for your stack. Pick one or more at step 1; each track determines which skills/plugins/rules are pre-checked in step 3.

Dev tracks

TrackStack
csr-supabaseVite + React + Supabase
csr-fastifyVite + React + Fastify
csr-fastapiVite + React + FastAPI
ssr-nextjsNext.js (App Router)
ssr-htmxHTMX + server-side
dataDuckDB + Polars + PySide6
fullunion of all dev tracks

Business tracks

TrackUse
executiveproposals, due diligence, decks, financial models
project-managementPM workflow + reviews
growth-marketinggrowth + content marketing

Meta

TrackUse
toolingBash + Markdown meta-projects (no app stack)

What gets installed per track

External assets are recommended automatically based on your track selection. Step 3 shows them pre-checked; you can toggle anything before install.

Frontend (csr-* / ssr-nextjs / full)

AssetWhatSource
react-best-practicesReact patterns and component guidelinesLokeshSakthivel
shadcn-uiRadix-based component copy + Tailwind themeshadcn (official)
web-design-guidelinesUX/UI best practicesLokeshSakthivel
impeccableUI design, critique, and visual review skillspbakaus
next-skills (ssr-nextjs only)Next.js App Router patternsvercel-labs

Backend (csr-* / ssr-* / full)

AssetWhatSource
railway-skillsRailway deploy + project/service/env managementRailway official
supabase-agent-skills (csr-supabase)Supabase agent skillsSupabase official
postgres-best-practices (csr-supabase)Postgres patternsSupabase official
supabase-cli (csr-supabase)Supabase CLI (supabase login for OAuth)npm
vercel-cli (csr-supabase)Vercel CLInpm
netlify-cli (csr-supabase)Netlify CLInpm

Data (data / full)

AssetWhatSource
polars-K-DensePolars — fast Rust DataFrame (pandas alternative)K-Dense-AI
dask-K-DenseDask — distributed processingK-Dense-AI
python-resource-managementMemory / CPU management patternswshobson
python-performance-optimizationProfiling + vectorizationwshobson
anthropic-data-pluginVisualization + SQL explorationAnthropic official

Business (executive / project-management / growth-marketing)

AssetWhatSourceTracks
anthropic-document-skillspptx / docx / xlsx / pdf authoringAnthropicexecutive · full
c-level-skills28 advisory skills (CEO/CFO/COO)claude-code-skillsexecutive · full
business-growth-skillsGrowth, finance, marketing playbooksclaude-code-skillsexecutive · full · growth-marketing
finance-skillsFinancial modelsclaude-code-skillsexecutive · full
pm-skillsPM workflowsclaude-code-skillsproject-management
product-skillsProduct discovery + deliveryclaude-code-skillsdev + PM
marketing-skills / content-creator / demand-gen / research-summarizerMarketing playbooksclaude-code-skillsgrowth-marketing

Dev Tools (all dev tracks)

AssetWhatSource
playwright-skillE2E test authoring with Playwrighttestdino-hq
find-skillsSearch and rank installed skillsvercel-labs
agent-browserBrowser automation CLI for agentsnpm
architecture-decision-recordADR authoringyonatangross
karpathy-coderPre-commit quality gate hookclaude-code-skills
product-skillsProduct engineeringclaude-code-skills

Workflow (opt-in — pick one or more at step 3)

Which one? See the Workflow curation guide — a vetted comparison of all 8 installable workflows (plus honest pointers to Spec Kit / Kiro, which we recommend but don't auto-install).

AssetWhatActivates
uzys-harness 6-Gate workflow/uzys:spec → /uzys:plan → /uzys:build → /uzys:test → /uzys:review → /uzys:ship with hook-enforced gatesThe 6-gate flow described below
superpowersAgentic skills framework, Anthropic official marketplaceobra/superpowers
ecc-plugin60 agents · 230 skills · 75 commandsaffaan-m
gsd-orchestratorOrchestration for large projectsget-shit-done-cc
openspecSpec-driven brownfield delta workflow (propose → apply → archive)Fission-AI
bmad-methodMulti-agent agile workflow (PM/Architect/Dev, 12+ agents)bmad-code-org
addy-agent-skills/spec /plan /build /test /review /ship /code-simplify skillsaddyosmani's workflow
wshobson-agentsMulti-agent orchestration workflows (full-stack/tdd/review), cross-CLIwshobson

Security & ECC (opt-in)

AssetWhatSource
trailofbits-skillsDifferential security reviewTrail of Bits
ecc-pluginECC plugin (project-scoped via prune-ecc.sh)affaan-m
ecc-pruneTrim ECC down to a curated set (4 agents + 8 skills + 3 commands)this project

Trust tiers

Every external asset carries a trust tier, shown as a badge in step 3:

  • ★ official — Anthropic-official marketplaces + this harness's own assets.
  • vetted — community assets with ≥ 1000 GitHub stars and active maintenance. Pre-checked when they match your track.
  • ⚠ experimental — under 1000 stars. Opt-in only (not pre-checked), listed at the bottom of each category.

Tiers inform, never block — you always review and choose what installs. Recommended assets (official/vetted matching your track) sort to the top.

How is "verified" backed up? See the Compatibility & verification matrix — install methods are checked against real registries/marketplaces, and the core workflow set is verified by real install in an isolated Docker container (not a static table). Trust tiers are auto-monitored for star-drift monthly.


6-Gate workflow (only if you opt in)

The 6-gate workflow is off by default. To enable it, check uzys-harness 6-Gate workflow at step 3 (or pass --with uzys-harness for non-interactive install).

/uzys:spec → /uzys:plan → /uzys:build → /uzys:test → /uzys:review → /uzys:ship
GatePurpose
/uzys:specDefine what you're building before code
/uzys:planDecompose the spec into small, verifiable tasks
/uzys:buildImplement incrementally
/uzys:testProve it works
/uzys:reviewMulti-perspective code/security review
/uzys:shipPre-launch checklist + deploy

Gates are enforced by hooks installed under .claude/hooks/. Skipping a gate fails the next one (exit code 2). If you didn't opt in, /uzys:* commands aren't installed and the hooks aren't active — you get the rest of the track's assets without the gating.


Scope

Step 4 asks where the install writes.

Project (default)Global (opt-in)
claude plugin--scope project (entries isolated by projectPath in installed_plugins.json)--scope user
npx skillsproject node_modules-g (user-level)
npm--save-dev (devDependency)-g
Codex (prompts / skills / config).codex/ in your project~/.codex/
Antigravity (skills / workflows).agents/ in your project~/.gemini/antigravity/
~/.claude/skills/ · ~/.codex/ · ~/.opencode/ · ~/.gemini/ · npm root -gnot touchedwritten per asset

Project scope keeps the install confined to your repo. Other projects on the same machine are unaffected.

Claude CLI itself writes plugin cache under ~/.claude/plugins/cache/ regardless of scope — but the metadata in installed_plugins.json isolates entries by projectPath, so other projects don't see them.


Uninstall

npx -y @uzysjung/agent-harness uninstall

Reads .claude/.harness-install.json (created during install) and reverses what was installed.

  • Project-scope assets: removed automatically (claude plugin uninstall --scope project, npm uninstall --save-dev, .codex/ cleanup, etc.).
  • Project root CLAUDE.md: removed only if it still matches the installed version (sha256); kept with a notice if you've edited it since install.
  • Global-scope assets: listed as advisory only — you remove them yourself. Uninstall never touches another project or your global config without explicit action.

Flags:

FlagWhat
--dry-runList reverse steps, change nothing
--keep-templatesRemove external assets but keep .claude/, .codex/, .opencode/

How it works

┌──────────────────────────────────────────────────────────┐
│  npx -y @uzysjung/agent-harness                         │
│         │                                                │
│         ▼                                                │
│  ┌─ 6-step wizard ──────────────────────────────────┐    │
│  │  Track(s) → CLI(s) → Items → Scope → Confirm    │    │
│  └──────────────────┬───────────────────────────────┘    │
│                     ▼                                    │
│  ┌─ Phase 1: Templates ─────────────────────────────┐    │
│  │  .claude/{rules,agents,hooks,commands,skills}    │    │
│  │  CLAUDE.md (merged) · .mcp.json                  │    │
│  └──────────────────┬───────────────────────────────┘    │
│                     ▼                                    │
│  ┌─ Phase 2: External assets ───────────────────────┐    │
│  │  claude plugin / npx skills / npm / shell-script │    │
│  │  Honors the scope chosen at step 4               │    │
│  └──────────────────┬───────────────────────────────┘    │
│                     ▼                                    │
│  ┌─ Phase 3: install log ───────────────────────────┐    │
│  │  .claude/.harness-install.json                   │    │
│  │  (drives `uninstall`)                            │    │
│  └──────────────────────────────────────────────────┘    │
└──────────────────────────────────────────────────────────┘

After install, a tooling + Claude project looks like:

your-project/
├── .claude/
│   ├── rules/          # coding conventions for your stack
│   ├── agents/         # subagent definitions
│   ├── commands/uzys/  # /uzys:* commands (only if you opted into uzys-harness)
│   ├── hooks/          # gate / pre-commit hooks
│   └── settings.json   # your existing one is backed up first
├── CLAUDE.md           # merged instructions (yours backed up if it differed)
└── .mcp.json           # MCP servers, merged with yours

CLI support

CLIStatus
Claude CodeFirst class — all assets and hooks
Codex (OpenAI)Skills + AGENTS.md rules for your stack. Global /uzys-* slash prompts (opt-in via ~/.codex/prompts/); project-level prompts await upstream support (#9848)
OpenCodeSkills + AGENTS.md integration
Antigravity (Google)Project: .agents/rules/ (context, always) + .agents/skills/ + .agents/workflows/ (6-Gate opt-in). Global (opt-in via --with-antigravity-global + --scope global): ~/.gemini/antigravity/{skills,global_workflows}/uzys-* (v26.66.0+)

Pick one or more at step 2.


Advanced


License

MIT.

Ver o original no GitHub

Bom para

Relacionados