A lightweight ChatGPT and Codex plugin that reduces unnecessary context, file reads, searches, tool output, and response length without sacrificing correctness or safety.
Distribution status: The public plugin package and submission materials are ready. OpenAI review and Universal Plugins Directory publication have not been completed yet.
token-efficientは、ChatGPTとCodexの作業品質を保ちながら、不要なトークン消費を減らすための軽量なskills-onlyプラグインです。外部サーバー、アカウント、認証、APIキーは必要ありません。
主に次の無駄を抑えます。
- ファイルやログの丸読み
- あてずっぽうな探索と同じ情報の再読
- 小さな変更でのファイル全体の書き直し
- 長すぎるツール出力、途中報告、最終回答
- 根拠のない正確な節約量の主張
必要な指示、影響範囲、安全確認、検証、エラー処理は省略しません。効率と品質が衝突する場合は、正確性・安全性・タスク完了を優先します。
暗黙発動は無効です。必要なタスクで明示的に選択してください。
- ChatGPTデスクトップ:
@token-efficient - Codex CLI / IDE:
$token-efficient、または/skillsから選択
Universal Plugins Directoryでの公開前は、standalone skillとして手動インストールできます。
~/.agents/skills/token-efficient/
├── SKILL.md
└── agents/
└── openai.yaml
Windows PowerShellの例:
git clone https://github.com/ta2ta2yo-sudo/token-efficient.git token-efficient-repo
New-Item -ItemType Directory -Force "$env:USERPROFILE\.agents\skills"
Copy-Item -Recurse -LiteralPath ".\token-efficient-repo\skills\token-efficient" -Destination "$env:USERPROFILE\.agents\skills\token-efficient"公開後はUniversal Plugins Directoryからインストールできる予定です。現在の提出状況は申請チェックリストで確認できます。
- コードや文書の調査・編集
- 大きなリポジトリ、ログ、データの確認
- 冗長になりやすい長時間タスク
- 品質を落とさず、読み込みと出力を絞りたい作業
token-efficient is a lightweight skills-only plugin for reducing unnecessary token usage while preserving the quality of work performed by ChatGPT and Codex. It requires no external server, account, authentication, or API key.
It focuses on avoiding:
- Reading entire files or logs when targeted ranges are sufficient
- Guess-driven exploration and repeated reads of unchanged evidence
- Rewriting whole files for small changes
- Excessively long tool output, progress updates, and final responses
- Unmeasured claims about exact token or cost savings
It does not skip required instructions, affected context, safety checks, verification, or error handling. When efficiency conflicts with quality, it prioritizes correctness, safety, and task completion.
Implicit invocation is disabled. Select the skill explicitly for tasks where you want it applied.
- ChatGPT desktop:
@token-efficient - Codex CLI / IDE:
$token-efficient, or select it through/skills
Until it is published in the Universal Plugins Directory, install it manually as a standalone skill:
~/.agents/skills/token-efficient/
├── SKILL.md
└── agents/
└── openai.yaml
Windows PowerShell example:
git clone https://github.com/ta2ta2yo-sudo/token-efficient.git token-efficient-repo
New-Item -ItemType Directory -Force "$env:USERPROFILE\.agents\skills"
Copy-Item -Recurse -LiteralPath ".\token-efficient-repo\skills\token-efficient" -Destination "$env:USERPROFILE\.agents\skills\token-efficient"After approval and publication, it is intended to be installable from the Universal Plugins Directory. Track the current state in the submission checklist.
- Investigating or editing code and documents
- Working with large repositories, logs, or datasets
- Long-running tasks that tend to accumulate context
- Tasks where reads and output should stay narrow without reducing quality
.
├── .codex-plugin/plugin.json
├── assets/
├── skills/token-efficient/
├── docs/
└── submission/
This plugin does not guarantee a specific token reduction. Actual usage depends on the task, model, tools, and required verification.
The skill follows the OpenAI agent skill format used by ChatGPT and Codex. See the OpenAI skill documentation and plugin packaging documentation.
MIT License. See LICENSE.
MITライセンスで公開しています。詳細はLICENSEを参照してください。