ccg-workflow

Orchestrates multi-model role agents through analyze, design, build, and review phases for code changes.

5.9k|445|Updated Jan 4, 2026
One-click install
npx skills add https://github.com/fengshao1227/ccg-workflow --skill ccg-workflow
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: ccg-workflow
Source: https://github.com/fengshao1227/ccg-workflow/tree/main/dsh-ccg/skills/ccg-workflow
Command: npx skills add https://github.com/fengshao1227/ccg-workflow --skill ccg-workflow

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Non-trivial code changes — whole features, unclear bugs, large refactors — often sprawl or fail when attempted in a single pass. This Skill structures the work by delegating phases to specialized role tools (ccg_analyze, ccg_design, ccg_build, ccg_debug, ccg_optimize, ccg_review, ccg_test), each running a child agent on its own model, so the work converges instead of sprawling.

Core Features & Use Cases

  • Four delegation modes: Direct, Standard, Deep, and Team modes let you match the workflow cost to the task size, from a one-line fix to a multi-agent parallel build.
  • Deterministic quality gates: Bundled verify-change, verify-quality, verify-security, verify-module, and gen-docs scripts scan diffs and modules before review so the reviewer focuses on judgement.
  • Team coordination: ccg_team hires persistent specialist agents with disjoint file ownership, while ccg_roster and ccg_remember track contracts, decisions, and conventions across sessions.
  • Use Case: When facing a bug whose cause is unknown, run ccg_debug to diagnose ranked hypotheses, then ccg_build to implement the fix, then verify-change and ccg_review to validate the result.

Quick Start

Ask the assistant to run the CCG workflow on a feature or bug, for example: use the ccg-workflow skill to analyze, design, build, and review the session refresh fix in src/auth/session.ts.

Frequently Asked Questions about ccg-workflow

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

FAQPage Schema
How do I run a multi-model code review workflow?

Delegate phases to role tools: run ccg_analyze or ccg_design first, then ccg_build, then the verify-* gate scripts, and finally ccg_review. Each role runs a child agent on its own model, so write a complete brief with goal, file anchors, constraints, and expected output shape.

When should I use Deep mode versus Team mode?

Deep mode buys certainty on one hard problem by asking several models about it, while Team mode buys throughput by running specialists in parallel on separate files. A hard question is Deep even if small; a wide job is Team even if each part is easy.

Should I delegate small fixes to the CCG role tools?

No. One-line fixes, renames, and questions answerable by reading a couple of files should be done directly. Delegation costs a round trip and a fresh context, so spend it only where the depth of a specialist agent is what you lack.

How do I prevent conflicts when running multiple agents in parallel?

Give every concurrent teammate a disjoint set of owned files and settle all shared contracts — signatures, schemas, event names — before hiring. Two agents editing one file lose each other's work, so sequence those tasks or split along file boundaries.

What do the verify-change and verify-quality gates check?

verify-change analyzes the diff and flags documentation that drifted from the code, while verify-quality checks complexity, duplication, naming, and length in complex or refactored modules. They are deterministic scanners whose findings inform your decision rather than verdicts.