Skill Templates para Claude Code
blamechris/skill-templates
Registry descentralizado de skills para automatizar workflows em múltiplos repos
Ver no GitHubRegistry centralizado de templates reutilizáveis de skills para workflows Claude Code. Permite que repositórios instalem skills sob demanda via cliente /skill, customizando templates inline sem deploy central ou fan-out.
Instalação
npx convez-ai blamechris/skill-templates1 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 processos de revisão e PR em múltiplos repositórios
- Distribuir skills customizáveis sem deploy central
- Manter versionamento e detecção de drift em skills instaladas
Veredito

Ferramenta robusta para times que gerenciam múltiplos repos e precisam de skills versionadas e auditáveis. Exige disciplina: sem source control automático, requer gestão manual de templates e validação via linter determinístico.
README
A tradução para o português deste README ainda está na fila.
skill-templates
A registry of reusable Claude Code skills — npm/brew for .claude/commands/*.md.
It is the canonical source of truth for review workflows, PR processes, and development
skills used across all managed repos.
Skills are not pushed out. Repos install them on demand via the /skill client,
and the invoking agent customizes each template inline for its repo. There is no central
fan-out, no deploy step, and no API key in this repo.
License: all rights reserved. This repo is public for visibility only — it is not open source. No permission to use, copy, or adapt these skills is granted by its availability. See NOTICE; contact @blamechris to request use.
Layout
skill-templates/
├── generic/ # Skill templates (repo-agnostic, with {{CUSTOMIZE: ...}} markers)
│ └── skill.md # the /skill client itself
├── registry.json # generated index: skill → template hash, description, guards
├── skill-guards.json # per-skill content guards (load-bearing markers)
├── scripts/
│ ├── build-index.sh # regenerates registry.json from generic/ + skill-guards.json
│ └── skill-lint.sh # mechanical gate: residual markers, attribution, version stamp, guards
└── docs/skill-profile-schema.md # the .claude/skill-profile.md spec
Using skills (consumer side)
In any managed repo, run /skill:
skill add <name>— resolve fromregistry.json→ fetchgeneric/<name>.md→ the agent fills{{CUSTOMIZE: ...}}markers from the repo'sCLAUDE.md+.claude/skill-profile.md+ code → self-validate → write.claude/commands/<name>.md(version-stamped) → lint the written file with the registry clone'sscripts/skill-lint.sh(a deterministic gate independent of the agent's judgment) → record in.claude/skills.lock. Consumers can run the same linter in a pre-commit hook or CI.skill list— show installed skills and their registry status.skill outdated— flag drift: version (template hash moved), profile (.claude/skill-profile.mdchanged), or corruption (aguardscheck fails).skill update [name]— re-install drifted skills.skill remove <name>— delete an installed skill.
Install-on-miss is a rule, not automatic tooling: per the global CLAUDE.md, if /X
is requested but absent from .claude/commands/, the agent runs skill add X first, then
invokes it.
Each consuming repo carries
.claude/skill-profile.md— the repo's customization profile (tech stack, CI commands, branch conventions, per-skill overrides). Seedocs/skill-profile-schema.md..claude/skills.lock— which skills are installed, at what template hash + profile hash..claude/commands/*.md— the installed, customized skills (version-stamped).
Maintaining the registry
- Edit a template in
generic/. - Commit.
- Run
./scripts/build-index.shto refreshregistry.json(hashes, descriptions, guards).
Consumers pick up the change on their next skill update. There is no deploy step.
Each installed skill carries a version stamp so drift is detectable:
<!-- skill-templates: <name> <hash> <date> -->
History
This repo previously used a push-deploy model (deploy.sh + deploy.conf +
a GitHub Action that fanned out a full N-repo × M-skill re-customization on every change).
That was retired in favor of the pull-based registry — see #68 (epic) and #75 (cleanup).
Do not reintroduce a push trigger, deploy.sh, or deploy.conf.
Attribution
The repo owner is the sole author of all work here. No Co-Authored-By trailers, no
"Generated with Claude" or any AI/agent attribution in commits, PRs, issues, or docs.