Reusable AI-assisted development workflows, prompts, and operating patterns.
This repository collects practical workflows for using AI coding tools safely and efficiently. The focus is on planner/executor and manager/specialist patterns: capable planning agents handle judgment-heavy work while bounded agents handle exploration, implementation, debugging, review, and documentation.
- Core Hybrid Model Workflow for Agentic Coding — the default: GPT-5.6 Sol High manages the project, Claude Code (Sonnet 5, Opus 4.8, or Fable 5) performs repository work by default, and Codex handles targeted or fallback work.
- Hybrid Model Workflow for Agentic Coding — the broader four-lane model: ChatGPT in the browser manages the project and selects among Codex, Claude Code, OpenCode, and Gemini according to task fit.
- OpenCode Go Router Workflow: Manager and Specialist Agents — replaces a manual multi-model copy-and-paste process with a repo-local OpenCode Go manager that routes work to specialist agents. Use it when you already have mid-tier browser or coding-agent subscriptions but want more repository-coding capacity, token budget, and role-based routing without maintaining multiple separate API subscriptions.
For OpenCode Go setup, follow the workflow's recommended project layout, prompt patterns, permissions, and manual smoke test. Use a project-root opencode.json plus prompts/, or a global ~/.config/opencode/ setup, as described in that workflow.
- Core hybrid workflow: GPT-5.6 Sol High manages. Claude Code executes by default. Codex handles targeted or fallback work.
- Hybrid model workflow: ChatGPT manages and selects among four repository-executor lanes: Codex, Claude Code, OpenCode, and Gemini.
- OpenCode Go workflow: Manager routes. Specialists act within approved boundaries.
Across these workflows, start with read-only discovery and planning when risk is unclear, keep edits narrowly scoped, run relevant checks, and review the Git diff before committing.