Forkable, multi-LLM skill distribution for AI coding agents.
Fork it, fill it with your company's skills, and sync them into Claude Code, Codex, and more — all from one versioned repo.
A quiver holds your arrows. quivo holds your skills — the markdown-plus-frontmatter capability packages that drive modern coding agents — and delivers the same set to every engineer's tools.
It is built to be forked. You take this repo, strip the examples, drop in your own skills, and your company now has a single source of truth:
- One repo — skills live in one place, version-controlled, history-tracked.
- One version for everyone — engineers
quivo syncand all see the same skill set. - Many agents — the same
SKILL.mdinstalls into Claude Code, Codex CLI, and (via adapters) other LLM harnesses. Author once, run everywhere. - Company-customizable — inject your own policy and environment config at install time without rewriting each skill.
quivo는 AI 코딩 에이전트용 "스킬"을 한 레포에서 관리·배포하는 포크형 템플릿입니다. 이 레포를 포크해 회사 스킬로 채우면, 모든 엔지니어가 같은 버전의 스킬을 Claude Code·Codex 등에 설치해 쓸 수 있습니다.
Requirements: Python 3.10+, uv (curl -LsSf https://astral.sh/uv/install.sh | sh).
For step-by-step walkthroughs with CLI examples and local testing, see Guide / 한국어 길라잡이.
Install it (recommended) — puts a real quivo on your PATH, then use quivo anywhere:
uv tool install git+https://github.com/chordpli/quivo.git # or: pipx install git+https://github.com/chordpli/quivo.git
quivo initUpgrade the CLI later with quivo update (or uv tool install --force git+...).
Or just try it (zero install) — run straight from the repo:
uvx --from git+https://github.com/chordpli/quivo.git quivo initOr alias the uvx form — no install, re-resolves each run:
echo 'alias quivo="uvx --from git+https://github.com/chordpli/quivo.git quivo"' >> ~/.zshrc && source ~/.zshrc
quivo initquivo init installs the bundled example skills into the current project:
.claude/skills/<name>/for Claude Code.codex/prompts/<name>.md+.codex/scripts/<name>/for Codex CLI
- Fork or clone
chordpli/quivo→my-company/quivo. For a private copy, clone and push to a new private repo — a GitHub fork of a public repo is always public. Throughout these steps,my-companyis a placeholder — replace it with your GitHub org or username. - Enable GitHub Actions in the fork (forks have Actions off by default).
- Point the CLI at your fork — edit one line in
quivo.yml:repo: my-company/quivo
- Add your skills under
skills/(use the bundledauthor-skillas a guide), bumpskills/VERSION, and push tomain. - Release — the
Release Skills Bundleworkflow buildsskills-bundle.tar.gzand publishes askills-v*GitHub Release in your repo. - Engineers install:
uvx --from git+https://github.com/my-company/quivo.git quivo init # or: quivo sync
The full step-by-step is in docs/quivo/fork-guide.md.
How a fork's skills reach each engineer's machine. Pick what fits your team.
| Mode | How | Private repo | Best for |
|---|---|---|---|
| A. Releases (default) | Fork builds skills-v* bundles; CLI downloads + caches |
Engineer provides a read token once | Many engineers, pinned immutable versions, sha256 integrity, offline cache |
| B. Local / clone | Clone the repo; install straight from disk | Uses your existing git auth (no token) | Maintainers, small teams |
# Mode B — install from a local checkout, no releases needed
export QUIVO_LOCAL_SKILLS=/path/to/quivo/skills
quivo init --agent bothFor private forks, releases work out of the box — set GH_TOKEN / GITHUB_TOKEN, or let the
CLI prompt once (it saves the token to ~/.quivo/token, mode 0600).
| Command | What it does |
|---|---|
quivo init |
Install skills into the current project (--agent claude|codex|both, --dir, --force, --release TAG, --no-policy) |
quivo sync |
Refresh installed skill content to the latest release |
quivo update |
Upgrade the quivo CLI itself to the latest cli-v* release |
quivo list |
List installed skills |
quivo doctor |
Check tooling, version, cache, and policy status |
quivo update upgrades the CLI; quivo sync refreshes the skills. Two tracks, two tags:
cli-v* for the CLI, skills-v* for the bundle.
Two install-time injection seams let one skill set behave differently per company — no need to fork the skill body:
.quivo/policy.md— appended to every installed skill (Prod procedures, permission rules, code/PR conventions). See the template in this repo..quivo/config.yml— per-engineer environment values (profiles, regions, IDs) that skills resolve at runtime. Copy from.quivo/config.example.yml.
Skip injection with quivo init --no-policy.
Kept intentionally minimal — this is a template, not a skill catalog.
| Skill | Purpose | Demonstrates |
|---|---|---|
author-skill |
Guide for writing new quivo skills to the standard | The authoring standard (docs/quivo/) |
aws-access |
Authenticate to AWS, verify identity, and serve as the entry point for AWS CLI work | setup.sh/setup.ps1 scripts, env-config (config › env › prompt) resolution, company-extension seam |
ripple |
Detect side effects of a change that tests miss (scan → clarify → resolve → check) | A self-contained, harness-agnostic SDLC skill with structured outputs and a delta-anchored Iron Law |
Authoring standard lives in docs/quivo/: the constitution, skill template,
permission model, and adapter reference.
quivo's design draws on the open-source frameworks that pioneered authoring and distributing agent skills. Their strengths, weaknesses, and what quivo borrowed (or deliberately rejected) are written up in docs/quivo/reference/inspiration.md (deep comparison in skill-framework-analysis.md).
- spec-kit — GitHub's Spec-Driven Development toolkit; spec → plan → tasks → implement, 30+ agent targets
- superpowers — Jesse Vincent's agentic SDLC discipline skills (Iron Law, TDD, verification)
- oh-my-claudecode — multi-agent orchestration with evidence-based verification gates
- gstack — Garry Tan's "virtual engineering team" personas (anti-sycophancy, hard gates, template generation)
- Anthropic Agent Skills — the first-party
SKILL.mdstandard (engineering blog · docs) - Andrej Karpathy — the minimalist "every token earns its place" skill-authoring philosophy
- spec-kit-ripple — origin of the bundled
rippleskill's scan → clarify → resolve → check flow
- Python 3.10+
uv(recommended) or pip- Network access (GitHub Releases) or
QUIVO_LOCAL_SKILLSfor offline/local use