菜单
Discovers and recommends AI models via the OpenRouter API. Matches task requirements (cost, speed, capabilities, context length, modality) to available models. Use when the user needs to choose a model, compare models, validate model IDs, or when selecting a model for a new script.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Read and create Google Calendar events. View today's schedule, upcoming events, and create new events. Use when user asks about calendar, schedule, meetings, or availability.
Creates Claude Code primitives (skills, commands, agents) following official best practices. Use when user wants to add new capabilities, workflows, or specialists. Asks clarifying questions to determine the right primitive type.
Dynamic Apify actor selection and execution for research tasks. Searches Apify store to find best actors, prefers pay-per-result pricing. Use when needing web scraping, data extraction, or research automation.
Extract content from any video - transcripts (speech-to-text) and AI understanding (visual analysis). Supports YouTube, Loom, and local files. Use when user mentions transcript, transcribe, video analysis, understand video, get text from video, video summary, extract from video, watch video, screen recording analysis, YouTube transcript, Loom transcript, or whisper transcription.
FFmpeg utilities for video file manipulation. Convert formats (MP4, WebM) and extract audio tracks. Use when user mentions video conversion, format change, extract audio, mp4, webm, ffmpeg, or file format.
Create and update Google Docs from markdown files. Convert local markdown to cloud documents with formatting preserved. Use when user asks to create a Google Doc, convert markdown to Google Docs, or update an existing doc.
| name | model-scout |
| description | Discovers and recommends AI models via the OpenRouter API. Matches task requirements (cost, speed, capabilities, context length, modality) to available models. Use when the user needs to choose a model, compare models, validate model IDs, or when selecting a model for a new script. |
Discover and recommend AI models from OpenRouter based on task requirements.
This skill activates when user:
ai-models.md are still current| Operation | Description | Command |
|---|---|---|
| Recommend | Score and rank models for a task | recommend |
| Compare | Side-by-side comparison of specific models | compare |
| Audit | Validate and improve ai-models.md | audit |
# Recommend models for a task
uv run .claude/skills/model-scout/scripts/scout.py recommend \
--task "summarize 500 documents cheaply"
# With specific requirements
uv run .claude/skills/model-scout/scripts/scout.py recommend \
--task "agent with tool calling" \
--needs tools,reasoning \
--min-context 128000 \
--top 5
# Compare specific models
uv run .claude/skills/model-scout/scripts/scout.py compare \
--models anthropic/claude-sonnet-4 google/gemini-2.5-flash-preview openai/gpt-4o
# Audit ai-models.md for stale entries
uv run .claude/skills/model-scout/scripts/scout.py audit --suggest-updates
recommend| Argument | Type | Default | Description |
|---|---|---|---|
--task | str | required | Natural language task description |
--needs | str | none | Required capabilities (comma-separated): tools, reasoning, vision, structured_outputs, audio |
--min-context | int | 0 | Minimum context window in tokens |
--max-cost | float | none | Max combined cost per 1M tokens (prompt + completion USD) |
--top | int | 5 | Number of results to show |
--all-providers | flag | false | Remove core-three provider bias for full exploration |
--json | flag | false | Output as JSON |
compare| Argument | Type | Default | Description |
|---|---|---|---|
--models | str[] | required | 2-5 model IDs to compare |
--json | flag | false | Output as JSON |
audit| Argument | Type | Default | Description |
|---|---|---|---|
--suggest-updates | flag | false | Print a ready-to-paste replacement table |
--json | flag | false | Output as JSON |
Models are scored across 5 dimensions (0-100 each) with dynamic weights:
| Dimension | What | Default Weight |
|---|---|---|
| Provider tier | Core three (OpenAI/Google/Anthropic) = 100, others = 50 | 15% |
| Cost | Log-scale inverse of prompt+completion per 1M tokens | 25% |
| Context | Size relative to requirement | 15% |
| Capabilities | Required params met + bonus for extras | 20% |
| Quality | Name-based tier lookup (opus=95, sonnet=80, flash=65, etc.) | 25% |
Weights shift based on task signal words (e.g., "cheap" boosts cost weight to 35%).
| Variable | Description |
|---|---|
OPENROUTER_API_KEY | API key from OpenRouter |