Skip to content

SKILL.md

Latest commit

 

History

History
296 lines (243 loc) · 8.74 KB

SKILL.md

File metadata and controls

296 lines (243 loc) · 8.74 KB
Raw
name chip-models
description Use when documenting or configuring public OpenClaw model aliases and provider notes without exposing secrets

chip-models

Public reference for OpenClaw model aliases and provider/auth notes. Keep secrets out of git; use placeholders or local-only auth files.

Contract

Primary class: Verifier Secondary class: Governor

Use when: documenting, auditing, or preparing safe model alias/provider/auth guidance for OpenClaw without exposing secrets.

Do not use when: the user asks to make a live config change and the exact config subtree has not been verified yet.

Output contract:

  • alias or provider mapping used
  • whether the statement is documentation-only or live-config applicable
  • relevant provider/auth caveat
  • explicit note when a config/schema check is still required before editing

Verification gate: do not present a model/provider detail as live-config truth unless it is verified against the actual runtime/config path.

Quick Reference

Alias Full Model ID Provider Purpose
kimi kimi-coding/kimi-code Kimi Coding Main Kimi coding/thinking model
codex openai-codex/gpt-5.3-codex OpenAI Codex Code generation
gpti openai-codex/gpt-5.3-chat-latest OpenAI Codex GPT-5.3 Instant
gptm openai-codex/gpt-5.4-mini OpenAI Codex GPT-5.4 Mini
gptt openai-codex/gpt-5.5 OpenAI Codex GPT-5.5 Thinking
sonnet anthropic/claude-sonnet-4-6 Anthropic General reasoning
opus anthropic/claude-opus-4-6 Anthropic Most capable
mmfast minimax/MiniMax-M2.7-highspeed MiniMax Fast responses
Minimax minimax/MiniMax-M2.7 MiniMax Full MiniMax M2.7
trinity openrouter/arcee-ai/trinity-large-thinking OpenRouter Optional Trinity shortcut

Recommended Config

Use only alias keys in agents.defaults.models unless a model explicitly needs params.

{
  "agents": {
    "defaults": {
      "model": {
        "primary": "kimi-coding/kimi-code",
        "fallbacks": [
          "openai-codex/gpt-5.5",
          "anthropic/claude-sonnet-4-6",
          "anthropic/claude-opus-4-6",
          "minimax/MiniMax-M2.5-highspeed",
          "minimax/MiniMax-M2.5",
          "openai-codex/gpt-5.3-codex"
        ]
      },
      "models": {
        "kimi-coding/kimi-code": { "alias": "kimi" },
        "openai-codex/gpt-5.3-codex": { "alias": "codex" },
        "openai-codex/gpt-5.3-chat-latest": { "alias": "gpti" },
        "openai-codex/gpt-5.4-mini": {
          "alias": "gptm",
          "params": { "transport": "auto" }
        },
        "openai-codex/gpt-5.5": {
          "alias": "gptt",
          "params": { "transport": "auto" }
        },
        "anthropic/claude-sonnet-4-6": { "alias": "sonnet" },
        "anthropic/claude-opus-4-6": { "alias": "opus" },
        "minimax/MiniMax-M2.7-highspeed": { "alias": "mmfast" },
        "minimax/MiniMax-M2.7": { "alias": "Minimax" },
        "openrouter/arcee-ai/trinity-large-thinking": { "alias": "trinity" }
      }
    }
  }
}

Provider/Auth Notes

Kimi Coding (kimi-coding)

  • Model ref: kimi-coding/kimi-code
  • In OpenClaw this provider is Anthropic-compatible.
  • If you define the provider block manually, use:
    • baseUrl: "https://api.kimi.com/coding/v1"
    • api: "anthropic-messages"
  • Wire endpoint: POST https://api.kimi.com/coding/v1/messages
  • Auth: API key via local-only config or auth profile

Important distinction:

  • kimi-coding/... = Kimi Coding
  • moonshot/... = Moonshot API
  • kimi/... is not the canonical way to reference current OpenClaw Kimi models

OpenAI Codex (openai-codex)

  • Model refs: openai-codex/gpt-5.3-codex, openai-codex/gpt-5.3-chat-latest, openai-codex/gpt-5.4-mini, openai-codex/gpt-5.5
  • Recommended auth mode in OpenClaw: OAuth (openai-codex:default)
  • This is the ChatGPT subscription flow, not pay-as-you-go API billing.
  • Do not document this as a custom api.openai.com/v1/chat/completions provider.
  • On current OpenClaw builds, define the provider models below if gpti, gptm, or gptt show up as configured,missing.
{
  "models": {
    "providers": {
      "openai-codex": {
        "baseUrl": "https://chatgpt.com/backend-api",
        "api": "openai-codex-responses",
        "models": [
          {
            "id": "gpt-5.3-chat-latest",
            "name": "GPT-5.3 Instant",
            "reasoning": false,
            "input": ["text", "image"],
            "contextWindow": 266240,
            "maxTokens": 266240
          },
          {
            "id": "gpt-5.4-mini",
            "name": "GPT-5.4 Mini",
            "reasoning": true,
            "input": ["text", "image"],
            "contextWindow": 400000,
            "maxTokens": 400000
          },
          {
            "id": "gpt-5.5",
            "name": "GPT-5.5 Thinking",
            "reasoning": true,
            "input": ["text", "image"],
            "contextWindow": 400000,
            "maxTokens": 400000
          }
        ]
      }
    }
  },
  "auth": {
    "profiles": {
      "openai-codex:default": {
        "provider": "openai-codex",
        "mode": "oauth"
      }
    }
  }
}

Optional transport override:

{
  "agents": {
    "defaults": {
      "models": {
        "openai-codex/gpt-5.4-mini": {
          "params": { "transport": "auto" }
        },
        "openai-codex/gpt-5.5": {
          "params": { "transport": "auto" }
        }
      }
    }
  }
}

Allowed values: auto, websocket, sse

OpenRouter Trinity (/trinity)

Alias: trinity -> openrouter/arcee-ai/trinity-large-thinking

Use for optional Trinity access without changing the default model.

Documented setup example:

{
  "agents": {
    "defaults": {
      "models": {
        "openrouter/arcee-ai/trinity-large-thinking": {
          "alias": "trinity"
        }
      }
    }
  }
}

Auth: provide OPENROUTER_API_KEY via local-only gateway env/config, not in git.

GPT-5.3 Instant (/gpti)

Alias: gpti -> openai-codex/gpt-5.3-chat-latest

Use for general inference tasks when you want GPT-5.3 without Codex tuning.

GPT-5.4 Mini (/gptm)

Alias: gptm -> openai-codex/gpt-5.4-mini

Use for lower-latency, lower-cost OpenAI Codex workloads when you still want strong tool use and reasoning.

GPT-5.5 Thinking (/gptt)

Alias: gptt -> openai-codex/gpt-5.5

Use when you want longer reasoning in the same OAuth-backed OpenAI Codex flow.

Kimi Multi-Key Rotation

Use multiple auth profiles for the same provider (kimi-coding) and order them in auth.order.

{
  "auth": {
    "profiles": {
      "kimi-coding:key1": { "provider": "kimi-coding", "mode": "api_key" },
      "kimi-coding:key2": { "provider": "kimi-coding", "mode": "api_key" },
      "kimi-coding:key3": { "provider": "kimi-coding", "mode": "api_key" }
    },
    "order": {
      "kimi-coding": [
        "kimi-coding:key1",
        "kimi-coding:key2",
        "kimi-coding:key3"
      ]
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "primary": "kimi-coding/kimi-code",
        "fallbacks": ["minimax/MiniMax-M2.5"]
      }
    }
  }
}

API Quick Ref

Provider Model ref example Auth mode Notes
Kimi Coding kimi-coding/kimi-code API key Anthropic-compatible (/coding/v1/messages)
OpenAI Codex openai-codex/gpt-5.3-codex OAuth ChatGPT subscription flow
OpenAI GPT-5.3 openai-codex/gpt-5.3-chat-latest OAuth /gpti alias
OpenAI GPT-5.4 Mini openai-codex/gpt-5.4-mini OAuth /gptm alias
OpenAI GPT-5.5 openai-codex/gpt-5.5 OAuth /gptt alias
OpenRouter Trinity openrouter/arcee-ai/trinity-large-thinking API key /trinity alias
Moonshot moonshot/kimi-k2.5 API key Separate provider line from kimi-coding

Usage

openclaw models list
openclaw models status
openclaw models auth login --provider openai-codex
openclaw onboard --auth-choice kimi-code-api-key

In chat:

Use kimi for this task
Model: codex
Model: gpti
Model: gptm
Model: gptt
Model: trinity

Claim wording rule

Separate:

  • documented reference — what this skill documents as the intended alias/provider setup
  • runtime verified — what has been checked in the actual current config/runtime
  • requires config/schema check — likely valid, but not yet safe to present as current truth

Avoid collapsing documentation examples into claims about the current live config.

Safety Checklist

  • Keep auth-profiles.json, .env, .env.local, and *.key ignored by git.
  • Use placeholders only in docs.
  • Do not paste exported OAuth state, tokens, or cookies into commits, screenshots, or markdown.