Claude Skills & Cookbook
dotts-h/claude-skills
13 skills de metodologia + recipes de engenharia instaláveis e versionáveis
Ver no GitHubMarketplace pessoal de plugins Claude que oferece 13 skills de metodologia (docs, processo, QA, design) e um sistema de recipes para instalar pacotes de engenharia versionados e reutilizáveis em repositórios. Funciona em sessões locais e na web/cloud.
Instalação
npx convez-ai dotts-h/claude-skills1 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
Casos de uso
- Padronizar práticas de documentação e decisões arquiteturais em múltiplos repos
- Implementar TDD, auditorias de qualidade e hardening de testes com metodologia consolidada
- Gerenciar tech debt e processos de engenharia como código versionado
Veredito

Ferramenta sofisticada para times que querem codificar e distribuir metodologias de engenharia. A separação entre skills (efêmeros) e recipes (duráveis) é bem-pensada, mas demanda disciplina organizacional para ser efetiva.
README
A tradução para o português deste README ainda está na fila.
ori — my Claude Code marketplace
ori is my personal Claude Code plugin marketplace (served from this repo), so my
tooling is available everywhere — including Claude Code on the web/phone (cloud
sessions install declared plugins at session start). It hosts two plugins:
| Plugin | Install id | What it is |
|---|---|---|
skills | skills@ori | 13 reusable methodology skills (docs, process, QA, design). |
cookbook | cookbook@ori | Recipe system: install versioned doc-based engineering-process packages into any repo, update via a lock + 3-way merge, check conformance with doctors. |
Use it in a repo (and from the phone)
Add to the repo's committed .claude/settings.json:
{
"extraKnownMarketplaces": {
"ori": { "source": { "source": "github", "repo": "dotts-h/claude-skills" } }
},
"enabledPlugins": {
"skills@ori": true,
"cookbook@ori": true
}
}
enabledPluginsis an object keyed byplugin-id@marketplace-idwith boolean values — the array form (["skills@ori", ...]) is silently ignored by the settings schema, so the plugins never enable.
Commit + push. Any cloud session for that repo installs the plugins automatically at session start (GitHub is in the default network allowlist), provided this marketplace repo is public — the cloud container clones the marketplace source directly from github.com without credentials. Locally you can also add them interactively:
/plugin marketplace add dotts-h/claude-skills
/plugin install skills@ori
/plugin install cookbook@ori
The skills plugin
13 methodology skills, bundled:
| Group | Skills |
|---|---|
| Docs | recording-decisions · registering-contracts · maintaining-conventions · logging-learnings · mapping-codebases |
| Process | practicing-tdd · managing-tech-debt · improving-architecture · auditing-code-quality |
| QA | hardening-tests · authoring-tests · exploring-quality |
| Design | designing-ui-ux |
The cookbook plugin
Installable, versioned, doc-based engineering process packages. A recipe turns a hard-won process (a constitution, quality gates, an issue store, a dev loop, a release ritual, a contracts registry) into something you install into a repo, keep updated, and conformance-check — the way a package manager treats code.
| thing | lives where | lifetime | who benefits |
|---|---|---|---|
| plugin / skill | the agent host | per-session context; leaves no trace in the repo | the agent that loaded it |
| recipe | the target repo (docs, scripts, CI, .recipes/lock.json) | durable state, versioned + updatable | any agent (or human) that opens the repo, on any host |
Skills here are the operators; recipes are the payload. A repo that adopted
recipes works with no plugin installed at all — the docs and scripts are just there,
host-agnostic (both CLAUDE.md and AGENTS.md thin pointers).
Quickstart
# In the repo you want to bring under management (with cookbook@ori enabled)
/adopt-recipes
adopt-recipes inventories the repo, proposes a profile + tier + answers, confirms
with you, installs in dependency order (harvesting existing docs on brownfield —
never clobbering), writes .recipes/lock.json, and reports gaps. Later:
/update-recipes (semantic 3-way merge to new versions, via PR) and /recipe-doctor
(aggregate conformance report).
Without the plugin, everything is runnable by hand:
make evals # run all recipe evals (fixture repos in mktemp)
make lint # bash -n all scripts + secret/date scans
make doctor TARGET=/path/to/repo # aggregate doctors against a repo
plugins/cookbook/scripts/install-recipe.sh \
--recipe plugins/cookbook/recipes/core --target /path/to/repo --tier M
Catalog
| recipe | layer | what it installs |
|---|---|---|
core | 0 | CLAUDE.md + AGENTS.md thin pointers, CONVENTIONS constitution (doctrine), CONTEXT glossary; ADR log (M+); CODEMAP/RETROS/TECH_DEBT (L) |
quality | 1 | single-run CI (make/npm flavors) with docs-only paths-ignore skip, the self-enforcing workflow guard, coverage-floor parameter, opt-in fuzz-smoke workflow, REGRESSIONS register (M+) |
loop | 1 | the dev loop playbook + start-fresh.sh (verified base) + next-issue.sh (dependency-aware picker) |
issues | 1 | hybrid issue store: markdown source of truth + GitHub mirror; INDEX + frontmatter format; epics via group: |
release | 1 | tag-driven release workflow with verified version resolution; SemVer + verify-after playbook |
contracts | 1 | CONTRACTS registry with Provides/Consumes; fleet constellation.yaml + fleet-doctor.sh cross-check |
Bindings (Layer 2, adapter docs — not recipes): frontend, api, qa,
services — each fills the slots Layers 0/1 leave open for that repo shape.
Profiles (recipes × tier × bindings × prefilled answers): app-full,
library, mini-api, mini-fe-components, mini-functions, greenfield-lean.
Tiers: S (mini repo: ~4 lean files, decisions inline in CONVENTIONS) · M (+ ADR log, REGRESSIONS, issues store, loop) · L (+ CODEMAP wiring, RETROS, TECH_DEBT).
Cookbook docs
- docs/recipes/SPEC.md — exact schemas:
recipe.yaml,lock.json, profiles,constellation.yaml, the doctor exit-code contract,min_model. - docs/recipes/DESIGN.md — why it's built this way.
- docs/recipes/RESEARCH.md — the original deep-dive research notes that governed the design.
Layout
.claude-plugin/marketplace.json # the ori catalog (both plugins)
plugins/skills/
.claude-plugin/plugin.json # the skills plugin manifest
skills/<skill>/SKILL.md # one dir per skill (+ references/, scripts/)
plugins/cookbook/
.claude-plugin/plugin.json # the cookbook plugin manifest
skills/<skill>/SKILL.md # adopt-recipes, update-recipes, recipe-doctor
recipes/<recipe>/ # recipe.yaml + templates + evals + doctor
profiles/ bindings/ scripts/
Makefile # cookbook recipe gates: make lint / make evals
CI
.github/workflows/ci.yml runs the cookbook recipe gates on every push/PR:
make lint (bash -n + secret/date scans) and make evals (every recipe's evals in
mktemp fixtures; needs bash + python3 + git).