Jim McMillan jimmc414

Build the Session Factory in your repo: a prompt for Claude Code

What this is. One prompt. Paste it into a Claude Code session opened at the root of your repository and it builds the process described in the thread: a session contract (CLAUDE.md), a stage map with exit gates, a frozen-plan template and the first frozen plan, a session ledger, a durable open-items register, decision records, a gate ladder with one pass string, a session-start signals probe, the continue, freeze, verify and queue commands, and an unattended queue runner that launches sessions overnight and judges each by its deliverable. It was condensed from a process specification written after about 500 autonomous sessions on one long-running build; every MUST in it has a recorded failure behind it, and the skeletons were dry-run in three toy repositories before publishing.

How to use it.

Fill in section 0 (nine facts about your repo and five knobs with defaults in brackets). Leave STOP-AND-ASK as (draft it) if you want th

Claude Code model-downgrade guard

A hook that halts a Claude Code session when the model serving it changes mid-session — for example when Fable 5 falls back to Opus 4.8 because credits ran out or a safety classifier flagged a request.

Without this, a session can silently continue for tens of minutes on a model you did not choose. In the case that prompted this, a session ran 309 messages on Fable 5, switched at the 310th, and continued for 91 more messages over 30 minutes before anyone noticed.

	# Agent Toolkit: Comprehensive Tool Inventory

	Status: Curated Reference Document (v9 - added RE category)
	Date: 2026-02-04
	Purpose: Catalog of agent-friendly CLI tools optimized for AI agent use
	Total Tools: ~275 (after removing 255 unsuitable tools, added Category 36: Binary Analysis & RE)

	---

	## Table of Contents

	# Spec: tmux Agent Panes

	> Fork Claude agents into observable tmux panes

	## Overview

	This skill enables Claude to spawn parallel agents, each working in their own tmux pane. The user can observe all agents working simultaneously in a single terminal window.

	```
	┌─────────────────────┬─────────────────────┬─────────────────────┐

	# Claude Code Skills Reference

	Version: 2.5
	Claude Code Version: 2.1.7 (fully tested)
	Date: 2026-01-15
	Methodology: Systematic testing of 167 skill configurations + diagnostic introspection

	---

	## Executive Summary

	# Analysis of "Verbalized Sampling: How to Mitigate Mode Collapse and Unlock LLM Diversity"

	## Response 1 (Probability: 0.23)

	### 1. Core Claim Contradicting Common Assumptions (Section 3)
	The paper argues mode collapse is fundamentally a data-level problem rather than an algorithmic one. Existing work attributed mode collapse to inadequate reward models or majority-favoring optimization. The authors prove typicality bias (α = 0.57±0.07, p<10^-14 in HELPSTEER dataset) exists in preference data itself, meaning "even with a perfect reward model and optimization process, inherent bias within preference datasets may still drive mode collapse."

	This matters because it implies you can't fix mode collapse by improving RLHF algorithms alone - the preference collection process itself needs reconsideration.

	### 2. Magnitude Finding (Section 5, Figure 4)

	# 🎨 Complete Visual UI Development Guide for Claude Code

	## 🚀 The Game-Changing Discovery

	Claude Code can SEE screenshots! This completely transforms UI development from blind coding to visual design with real-time feedback.

	### The Transform
	- Old Way: Write code → Hope it looks good → Fix bugs when reported
	- New Way: Build → SEE → Judge → Refine → Repeat until beautiful