Set up a repeatable workflow for delegating coding tasks to different AI models based on risk and difficulty.
Keep an auditable record of every delegated task's instructions, actions, and final result.
Run multiple AI coding agents in parallel without them overwriting each other's files.
Require independent review of high risk or hard to reverse changes before merging.
| yurudeveloper/explicit | c0urag1/break-risk-intel-skill | dw-dwain/ai-gen-company-stack | |
|---|---|---|---|
| Stars | 1 | 1 | 1 |
| Language | PowerShell | PowerShell | PowerShell |
| Setup difficulty | — | easy | easy |
| Complexity | — | 2/5 | 2/5 |
| Audience | developer | ops devops | developer |
Figures from each repo's GitHub metadata at analysis time.
This repository is not a piece of software in the usual sense. It is an operations guide, written as a CLAUDE.md file, that defines a workflow for coordinating several different AI models on a single coding project. It describes which project files to read first, which model to assign to which kind of task, where the work happens, what gets written down, who reviews it, and how progress is handed off between sessions. The guide sets up a lead orchestrator model that reads a fixed set of project documents in order: current progress, approved design, underlying concepts, and known gaps between the current implementation and the target. Based on the task, it assigns the work to one of several named models, ranging from one used for architecture and hard decisions down to a cheaper model reserved for routine, narrow, mechanical checks. Since the orchestrator cannot directly call these other models itself, it delegates through a thin wrapper script that records the exact instructions sent, invokes the chosen model, and saves the resulting output and final result as separate files. Every delegated task is required to produce three files: the exact input given to the worker model, a full log of what the worker did, and a short final result meant for the orchestrator to read without wading through the whole execution log. When several pieces of work happen at once, each one runs in its own isolated git worktree so parallel agents cannot overwrite each other's files. For risky or sensitive changes, such as anything security related, concurrency related, or hard to reverse, the guide calls for a separate reviewer model to independently check the work before it is accepted, rather than trusting the author's own report. Finally, the guide insists that a running progress file be updated immediately after any commit, merge, or design change, so that whichever session picks up the work next knows exactly where things stand. The project is mostly documentation and small PowerShell and bash helper scripts rather than an application with its own features.
A written operations guide that coordinates multiple AI models on one coding project, defining who works on what, how work is recorded, and how it is reviewed and handed off.
Mainly PowerShell. The stack also includes PowerShell, Bash.
Mainly developer.
This repo across BitVibe Labs
Don't trust strangers blindly. Verify against the repo.