adgaudes

SupervisedCoding — AI skill for Claude Code

AI community

Pi Agent extension for supervised coding: a supervisor (Pi) plans, delegates and verifies; workers (Claude through Claude Code, GPT through Pi's CLI) implement, with credit failover and per-repository.

How to install SupervisedCoding

This entry records only its repository, not the path inside it, so there is no exact command to give. Open adgaudes/SupervisedCoding and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What SupervisedCoding does

Alternatives in AI

Zotero MCP — It's a plugin extension in Zotero 1.1k ★
Fusion Plan — OMC interview (interactive) → 3-round Opus 4.8 + GPT-5.5 fusion deepening → concise .omc/plans/ output → revie 468 ★
T3 Pstack — pstack for T3 Code: poteto's pstack (via Michael Denyer's pstack-claude) with panels and fan-outs dispatched a 193 ★

README

SupervisedCoding

**Supervised multi-model coding for the [Pi](https://pi.dev) coding agent.** One model plans, delegates and accepts. Claude and GPT workers implement. Checks, independent reviews, credit failover and repository lessons run on their own.

![Pi extension](https://img.shields.io/badge/Pi-extension-6f42c1) ![Node.js ≥ 22.16](https://img.shields.io/badge/node-%E2%89%A5%2022.16-339933) ![Models: Claude · GPT](https://img.shields.io/badge/models-Claude%20%C2%B7%20GPT-555) [![License: MIT](https://img.shields.io/badge/license-MIT-blue)](LICENSE)

**English** · [Italiano](#italiano)

SupervisedCoding turns Pi into a small engineering team. The model you talk to becomes a **supervisor**: it explores the code, classifies the task, writes a precise implementation guide and makes the final call. **Workers** — Claude through Claude Code, GPT through Pi — make the change. The extension runs your tests before and after, sends regressions back to the same worker, gets the change reviewed by a different model family, and keeps going when a subscription runs out.

Its priorities, in order:

correct, well-made code with as few defects as possible;
the fewest tokens that quality allows — fewer in total, not the same amount moved to another model;
the right model and effort for each situation;
no interruptions when a model runs out of credits: the next one takes over;
speed comes last.

Delegation is a means, not the goal. The supervisor delegates when handing work to a disposable context costs less than holding that code in its own, and makes the change itself when it does not. [When it pays off](#when-it-pays-off) has the numbers.

Features

Any model, any role. Claude and GPT models can supervise, implement and review. Roles follow the task, never the model family. Claude workers run in Claude Code; GPT workers and any other model configured in Pi run through Pi's own CLI.
Structured delegation. The supervisor classifies each task