convez.ai
Skill
Naimul-islam-bd

token-efficiency

Naimul-islam-bd/token-efficiency

Ver no GitHub
InicialPluginScore · de 0 a 100

A Claude skill that cuts token waste lean replies, smart file reading, zero redundant work

Instalação

$npx convez-ai Naimul-islam-bd/token-efficiency

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
v1.0.0
Issues abertas
0
Contribuidores
1

README

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

token-efficiency

A Claude skill that makes every conversation cheaper.

I built this because I kept burning through my usage limit halfway through the day. Watching the responses closely, I realised most of the tokens weren't going into the actual answers — they were going into padding: restating my question back to me, re-reading files it had already read, reprinting a whole 400-line script to change six lines, ending every reply with a summary of the reply.

This skill is a set of rules that cuts that waste. Same quality of work, fewer tokens spent getting there. In my own use the difference is very noticeable on long sessions and file-heavy work — I'm not going to put a fake percentage here, but if your sessions look anything like mine (spreadsheets, field data, long documents, multi-step tasks), you'll feel it.

What it actually does

When the skill is active, Claude:

  • answers first, skips the preamble and the closing recap
  • reads each file once and works from memory instead of re-reading
  • reads the part of a large file it needs, not the whole thing
  • shows you the changed lines, not the whole rewritten document
  • merges multiple shell/search calls into one
  • stops searching once the question is answered
  • keeps short questions to short answers
  • never pads with unsolicited alternatives and "you might also want to..."

And one important thing it does not do: it never shortens code you'll run, truncates exact values, or cuts detail you explicitly asked for. Correctness always wins over compression.

Install

Claude Code:

mkdir -p ~/.claude/skills
git clone https://github.com/Naimul-islam-bd/token-efficiency.git ~/.claude/skills/token-efficiency

That's it — it triggers automatically on new sessions.

Claude.ai: download token-efficiency.skill from the releases page and upload it in Settings → Capabilities → Skills.

Structure

token-efficiency/
├── SKILL.md                 # the rules
└── references/
    └── examples.md          # before/after examples for each rule

The skill itself follows its own advice: SKILL.md stays small, and the examples live in a separate file that's only loaded when needed.

A note on expectations

No skill can magically divide your token bill by ten — anyone claiming that is selling something. What this does is remove the avoidable waste, which in chatty, file-heavy workflows is a large share of total usage. Your savings will depend on how you work.

Issues and PRs welcome. If you find a wasteful pattern these rules miss, open an issue — that's exactly the kind of thing I want to add.

License

MIT

Ver o original no GitHub

Relacionados