convez.ai
SkillProdutividadeIntermediário
abhishekray07

CC Wrapped

abhishekray07/cc-wrapped

Seu 'Wrapped' local: veja onde seus tokens e tempo realmente foram

Ver no GitHub
InicialPluginScore · de 0 a 100

Skill para Claude Code que analisa seus logs locais (~/.claude) e gera um relatório estilo Spotify mostrando onde seus tokens, tempo e trabalho realmente foram. Inclui classificação automática de tipos de trabalho, PRs mergeados rastreados e estatísticas de custo por modelo — tudo rodando localmente, sem deixar sua máquina.

Instalação

$npx convez-ai abhishekray07/cc-wrapped

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
0
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Entender padrão de uso e distribuição de tokens entre projetos
  • Rastrear tipo de trabalho realizado (features, bugs, docs, research)
  • Visualizar PRs mergeados e documentação criada em período
  • Gerar recaps animados com Remotion para compartilhar

Veredito

Ferramenta prática e bem pensada para quem vive no Claude Code. A análise de tipos de trabalho por heurística (sem LLM extra) e integração com GitHub PR é inteligente. Recomendado para quem quer visibilidade real de padrões de uso sem sair do terminal.

README

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

cc-wrapped

A local "Wrapped" for Claude Code. It reads the session logs Claude Code already writes to ~/.claude/projects and shows you where your week actually went — what kind of work, what shipped, what you were doing, and some fun stats.

No API key, no cloud, nothing leaves your machine.

cc-wrapped

Why

I do almost everything in Claude Code now — code, debugging, blog posts, research. I got curious where the time actually goes, and I didn't want to leave the terminal to find out. So, /cc-wrapped.

(Turns out only ~60% of my week was actually code.)

Install

It's a Claude Code skill. Clone it into your skills directory and it just works:

git clone https://github.com/abhishekray07/cc-wrapped ~/.claude/skills/cc-wrapped

Then in any Claude Code session:

/cc-wrapped          # last 7 days
/cc-wrapped 30       # last 30 days

Or run it straight as a CLI:

python3 ~/.claude/skills/cc-wrapped/cc_wrapped.py --days 7

Flags: --days N, --no-gh (skip the GitHub PR lookup), --json (raw stats), --top N.

Pure-stdlib Python 3, nothing to install. The "what shipped" section uses gh (the GitHub CLI) if it's authenticated; everything else works without it.

What you get

  • Where your tokens went — cost by project and model, real tokens vs cache, a daily-spend sparkline.
  • What kind of work — every session classified: building / reviewing / docs / research / tests / …
  • What shipped — your merged PRs (matched back to the session that opened them) and the docs you wrote.
  • What you were doing — your priciest sessions, each with the actual task.
  • Fun stats — longest session, busiest day, top tools, research vs ship.

How it works

Claude Code writes every session to ~/.claude/projects/<project>/<session>.jsonl — one file per session, one JSON event per line (your prompts, each assistant turn with its token usage, every tool call). cc-wrapped reads those.

  • Work type is a heuristic, no extra LLM call. It weighs what a session actually did (files edited, lines changed, tool mix, slash commands) over what the first prompt said, so a "how do I test this" that turns into 3,000 lines gets called a feature, not a test.
  • What shipped comes from gh search prs --author=@me, matched to the session whose transcript opened each PR, plus the substantial markdown files you wrote.
  • Cost is computed the way ccusage does it — dedup the API responses Claude Code replays across forked/resumed sessions, and pull per-model rates from LiteLLM's pricing DB (cached locally). If you just want raw token cost, use ccusage; this adds the "what was the work" layer on top.

It's a heuristic, so it'll file the occasional 4-hour rabbit hole under "feature."

The video (optional)

The animated recap in media/demo.mp4 is made with Remotion: cc_wrapped.py --json dumps the stats, you drop the numbers into a React composition, and one command renders it. See video/.

Tests

python3 test_cc_wrapped.py

License

MIT

Ver o original no GitHub

Bom para

Relacionados