菜单
Use when the user asks to run a task on a specific model, compare model outputs, or says 'run this on GPT/Gemini/Sonnet'. Routes tasks to any model via claude-code-router. Usage: /ccr-run [model] <task>
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
Convene the 18-member Council of High Intelligence — simulate multi-persona deliberation from Aristotle, Feynman, Torvalds, Taleb, Kahneman, and 13 others, then synthesize into ONE verdict. Use when facing hard decisions, design tradeoffs, strategy questions, or when you want diverse adversarial perspectives on a problem. Trigger on "/council", "ask the council", "what would the council say", "multi-perspective analysis". Supports --quick (fast), --duo (dialectic pair), --triad <domain>, --full (all 18), --members a,b,c.
Finds viable product, business, or project ideas in any domain by generating candidates, rigorously validating each against real competitors and market data, and killing weak ideas with documented reasons. Two modes — fast batch generate-and-kill (default), or `--deep` DFS that branches a single space into opportunity tuples with codex cross-attack, mandatory steelman, and a VC-gauntlet fundability gate. Use when the user asks to brainstorm ideas, find market gaps, identify business opportunities, figure out what to build, come up with side project ideas, find underserved niches, or (with --deep) "find the real opportunities in X", "exhaustively explore X", "keep digging until you exhaust X". Works for SaaS, open source, MCP servers, apps, tools, APIs, or any product category.
Stress-test a startup/product proposal like a top-tier VC partner meeting — reality-checked against real market data, with an independent codex devil's-advocate bear case — then iterate hardened rewrites to a GO / CONDITIONAL-GO / NO-GO verdict. Use when the user says "vc review", "is this fundable", "would a VC fund this", "tear apart my startup idea", "pitch critique", "investor lens", "gauntlet this", "red-team my startup", "poke holes in my idea". Also the fundability gate invoked by opportunity-dfs §8.
The single entry point for ALL design and specification work. Use when designing, specifying, architecting, writing a spec, RFC, design doc, or drafting a design for any system, feature, product, protocol, game, or workflow. Trigger phrases include "design this", "architect this", "design spec", "design review", "write a spec", "technical spec", "RFC", "API design", "data model", and "implementation spec". Default mode: adversarial stress-testing with parallel critics. Use `--spec` for lightweight single-session spec document generation without critics.
Use when researching, investigating, or exploring a topic systematically with orthogonal multi-dimensional coverage and source-quality tiers. Trigger phrases include "research this deeply", "deep research on", "investigate this topic thoroughly", "explore this topic", "systematic research", "multi-dimensional research", "comprehensive research", "cover all angles of", "thorough research on", "deep dive into (research)", "exhaustive research". Spawns parallel agents across WHO/WHAT/HOW/WHERE/WHEN/WHY/LIMITS with risk-stratified spot-checking. Bounded by a user-controlled round budget with honest coverage reporting on what was and wasn't covered.
Use when the user says "autopilot", "build this", "make it", "ship it", "just do it", "do it", "implement this", or gives a vague idea, spec, task list, or bug fix and expects autonomous execution. Handles decomposition, agent coordination, and verification automatically. "build me end to end", "go from idea to code", "do everything", "run this autonomously". Follows the Spec-Driven Development workflow: specify → clarify → plan → tasks → analyze → implement.
| name | ccr-run |
| description | Use when the user asks to run a task on a specific model, compare model outputs, or says 'run this on GPT/Gemini/Sonnet'. Routes tasks to any model via claude-code-router. Usage: /ccr-run [model] <task> |
| category | tool |
| input_types | ["task","question","code-path"] |
| output_types | ["code"] |
| complexity | moderate |
| cost_profile | low |
| maturity | beta |
| metadata_source | inferred |
You are a model router. Parse the user's input to determine the model and task.
Requires claude-code-router running (ccr start). See /ccr-models to list available models.
Check if the FIRST word matches a shortcut or full model ID. If it does, use that model and treat the rest as the task. If it doesn't match anything below, treat the ENTIRE input as the task and use AskUserQuestion to let the user pick a model.
See _shared/ccr-shortcuts.md for the model shortcut table.
If no model is identified, use AskUserQuestion with these options:
Spawn an Agent. Place this tag at the very START of the agent's prompt (before anything else):
{resolved model ID}
Then include the user's task as the rest of the prompt. The tag is stripped by the router before the model sees it.
