commit-all
ahpxex/commit-all
Commit e push em todos seus repos ativos com um clique
Ver no GitHubSkill que comita e faz push de todos os repositórios git que você trabalhou recentemente em um único comando. Descobre repos automaticamente a partir dos registros de projetos abertos em Claude Code, Codex, Cursor e outros agentes, sem varredura do filesystem.
Instalação
npx convez-ai ahpxex/commit-all1 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
- Sincronizar múltiplos projetos locais de uma vez após sessão de trabalho
- Automatizar limpeza de branches locais não sincronizadas
- Integrar com fluxo de trabalho multi-agente (Claude Code, Cursor, Codex)
Veredito

Ferramenta bem pensada para quem trabalha com múltiplos repos. O design de segurança (confirmação obrigatória, proteção contra push de branches atrasadas) é responsável. Limitação: só funciona com repos que você abriu em agentes suportados — não substitui uma varredura completa do filesystem.
README
A tradução para o português deste README ainda está na fila.
commit-all
A coding-agent skill that commits — and pushes where safe — every git repo you've recently worked in, in one shot.
The hard part of "commit everything" isn't committing; it's finding the repos.
commit-all doesn't scan your filesystem. Instead it reads your coding agents' own
records of recently-opened projects and uses those as the index of active work. So
coverage automatically tracks where you actually code, with zero configuration.
Works as a Skill across any agent the skills CLI supports
(Claude Code, Codex, Cursor, Gemini CLI, Copilot, and 70+ more).
How it works
agent project records ──► resolve to git root ──► classify ──► commit / push
(Claude Code, Codex, git rev-parse git status by bucket
Cursor; deduped union) --show-toplevel --porcelain=v2
-
Collect candidate paths from every supported agent (union, deduped):
Agent Source Claude Code ~/.claude.json→.projectskeysCodex ~/.codex/config.toml[projects."…"]+~/.codex/sessions/**/*.jsonlcwdCursor …/globalStorage/state.vscdb→history.recentlyOpenedPathsListThe script reads each agent's config directly, so it sees all agents' projects no matter which agent invokes the skill.
-
Resolve each path to its git toplevel (
git rev-parse --show-toplevel). Non-repos, parent directories, and duplicates drop out automatically. -
Classify each repo (
git status --porcelain=v2 --branchplus remote/upstream probing) and sort into buckets:Bucket Meaning Action readydirty or ahead, has a push target commit + push commit_onlyno remote, or branch not on remote yet commit only blockedbehind the remote commit locally, never auto-push clean nothing to commit, already synced skipped silently -
Commit each actionable repo with a message generated from its diff, then push per bucket. Behind repos are reported, not force-pushed.
Tradeoff: a repo you've never opened in a supported agent won't be discovered — intentional, since it keeps the list to active work.
Install
Via the skills CLI — it installs into every coding agent it
detects on your machine:
# From a published GitHub repo (shorthand, full URL, or any git URL)
npx skills add <owner>/<repo>
# From a local checkout
npx skills add /path/to/commit-all
# Manage
npx skills list
npx skills remove commit-all
Requirements: git and bun (discovery is a single TypeScript
file using only Bun built-ins — no bun install needed).
Usage
In any installed agent:
/commit-all
or just ask it to "commit all my repos" / "提交所有仓库". The skill will:
- Run discovery and show you the actionable repos grouped by bucket.
- Ask for confirmation before writing anything.
- Commit each repo with a generated message and push where safe.
- Print a per-repo summary (committed / pushed / skipped / blocked).
You can also run discovery standalone to preview, without committing:
bun discover.ts --pretty # human-readable
bun discover.ts # JSON
Safety
- Confirmation required before any commit or push — this writes across many repos and is hard to undo.
- Behind-remote repos are never auto-pushed. They're reported so you can
pull --rebase(or merge) yourself first. No automatic pull/rebase/force. - Secret guard: if a staged diff adds a
.env, key file, token, or credential, the skill pauses and asks instead of committing it. - Upstreams aren't created silently: pushing a brand-new branch with
-uis confirmed first.
Extending discovery to another agent
Discovery sources live in a registry in discover.ts:
const SOURCES: Record<string, Source> = {
"claude-code": srcClaudeCode,
codex: srcCodex,
cursor: srcCursor,
};
Add a collector that returns an array of candidate paths (and fails closed —
returns [] if that agent isn't installed or anything goes wrong), then register it:
function srcMyAgent(): string[] {
// ...
return paths;
}
SOURCES["myagent"] = srcMyAgent;
Everything downstream (resolve, classify, bucket) is agent-agnostic and needs no changes.
Files
| File | Purpose |
|---|---|
SKILL.md | Skill definition + workflow the agent follows |
discover.ts | Discovery, resolution, and classification (Bun, zero deps) |
README.md | This file |