skills.rest
amitpo23
amitpo23COMMUNITY
Explore all skills by this author →
GitHub

council

Orchestrates multi-model code review and implementation by routing subtasks to specialized LLM experts.

Updated Jan 9, 2026
Search Tags:#tdd#code review#debugging#routing#orchestration#multi-model#llm council
One-click install
npx skills add https://github.com/amitpo23/cfo --skill council-amitpo23
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: council
Source: https://github.com/amitpo23/cfo/tree/main/.claude/skills/council
Command: npx skills add https://github.com/amitpo23/cfo --skill council-amitpo23

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Complex coding tasks often exceed what a single model handles well, and ad-hoc multi-model consultation produces conflicting opinions with no clear resolution. This Skill provides a structured routing protocol that decomposes complex tasks, assigns subtasks to specialized models (architecture, planning, implementation, QA), and synthesizes one verified, actionable answer. ## Core Features & Use Cases - Model Routing: Assigns deep codebase analysis to GPT 5.6 Sol, planning and critique to Opus 5, implementation to Sonnet 5, and post-implementation QA to GPT 5.5. - Staged Protocol: Enforces a plan → implement (TDD) → review → synthesize workflow with conflict resolution verified against the actual code. - Cost Discipline: Skips the council entirely for simple tasks and runs independent experts in parallel with timeouts and fallback rules. - Use Case: When facing a cross-layer bug or an architecture decision with trade-offs, convene the council to get competing expert analyses, then receive a single decided plan implemented by the Sonnet 5 contractor and audited on the real git diff. ## Quick Start Ask the council to analyze and resolve a complex cross-layer bug or architecture decision in this repository.

Frequently Asked Questions about council

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run a multi-model code review on a git diff?▼

Pass the git diff inside the prompt to GPT 5.5 via codex exec for bug, edge-case, and security review, and to Opus 5 via the claude CLI for simplification critique. The orchestrator then verifies each finding against the actual code before applying fixes.

When should I use a multi-model council instead of a single model?▼

Use the council only for complex tasks such as architecture decisions, cross-layer bugs, or choices with real trade-offs. Simple fixes, factual questions, or single-file changes should be handled directly, since every expert call costs real tokens.

How are conflicting opinions between AI models resolved?▼

The orchestrator decides by verifying each claim against the actual codebase following a verify-first doctrine. Unverified expert claims are treated as hypotheses, and the final output is one clear, actionable answer rather than a list of competing opinions.

What CLI tools are required to run the council experts?▼

GPT 5.6 Sol and GPT 5.5 run through codex exec with read-only sandbox, requiring codex-cli version 0.145.0 or higher. Opus 5 runs through the claude CLI with the prompt piped via stdin, and Sonnet 5 runs through the Agent tool inside the session.

What happens when a codex exec expert hangs or times out?▼

Broad repo-investigation prompts can stall codex exec for hours, so prompts should be focused with a closed file list and a 10-minute timeout. After two failures, continue without that expert and note the omission in the report.

Related Skills

Explore More Software Engineering Skills→
meta-pytorchmeta-pytorch
2.5k

write-tests

Generate failing test scaffolds from todos or requirements for TDD workflows.

Official
Intermediate
cabeznocabezno

test-driven-development

Enforce test-first development with failing tests before production code.

Community
Advanced
klu-devklu-dev
2

terminal-ops

Execute repository commands and verify outcomes with git state management.

Community
Intermediate
telefrektelefrek

feature-quick

Implement small code changes with TDD and minimal setup.

Official
Basic