zeroapi
por dorukardahan
Coding Agents & IDEsResumo
Route tasks to the best AI model across paid subscriptions (Claude, ChatGPT, Codex, Gemini, Kimi) via OpenClaw.
Descrição
O que esta skill faz
ZeroAPI routes tasks to the optimal AI model across your active subscriptions using a 9-step classification algorithm. Each task type maps to the model with the strongest benchmark: Codex for code and math, Gemini Pro for large context and research, Kimi for multi-agent orchestration, Opus for reasoning and judgment.
Using one model for everything leaves benchmark advantages unused; ZeroAPI picks the specialist with the strongest score for each task type.
Quando usá-la
- Automatically routing a math proof to Codex instead of Opus
- Analyzing a 200-page PDF using Gemini Pro's 1M context window
- Splitting code writing to Codex and code review to Opus in one session
- Running low-latency health checks with Flash-Lite's 0.23s TTFT
- Coordinating a 3-agent research, write, and review pipeline via Kimi
Exemplo de fluxo de trabalho
User asks: Analyze this 200-page PDF and write a Python parser for its data format
- 1
Estimate token count: a 200-page PDF exceeds 100k tokens, Step 1 matches
- 2
Delegate to RESEARCH tier: /agent gemini-researcher Analyze the attached PDF and identify the data format structure
- 3
Receive structured analysis identifying the format fields and layout
- 4
Delegate to CODE tier: /agent codex Write a Python parser for the format described: <summary from previous step>
- 5
Merge sub-agent outputs into a single response with analysis and working parser code
Structured format analysis from Gemini Pro plus a working Python parser from Codex, combined into one response
Requisitos
- Claude Max subscription (5x or 20x plan)
- ChatGPT Plus or Pro subscription (optional, required for Codex access)
- Google Gemini Advanced subscription (optional, required for Flash and Pro tiers)
- Kimi API key (optional, required for orchestration tier)
Instalação
clawhub install dorukardahan/zeroapiExecute o comando da CLI ou cole o prompt no Claude Code, Cursor, Codex ou no agente de código que você usa.