Claude-skill-registry divergence-control

Keep multiple instances aligned while allowing productive variance

install

source · Clone the upstream repo

git clone https://github.com/majiayu000/claude-skill-registry

Claude Code · Install into ~/.claude/skills/

T=$(mktemp -d) && git clone --depth=1 https://github.com/majiayu000/claude-skill-registry "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skills/data/divergence-control" ~/.claude/skills/majiayu000-claude-skill-registry-divergence-control && rm -rf "$T"

manifest: skills/data/divergence-control/SKILL.md

Divergence Control

Purpose

Instances naturally diverge as they think different thoughts. This skill manages that divergence:

Prevent wild deviation (instances completely disagreeing)
Allow productive variance (different approaches to same problem)
Maintain coherence (all instances solving related problems)

The Problem

Too much control: All instances think identically (no benefit) Too little control: Instances diverge so much they're solving different problems

Just right: Instances explore different solution paths while staying on the same problem.

Core Pattern

Instance 1: Path A ─┐
Instance 2: Path B ─┼─ Stay coherent
Instance 3: Path C ─┤  (same problem,
Instance 4: Path D ─┘   different approaches)

Key Features

Problem Anchoring - All instances address the same core question
Variance Measurement - How different ARE the outputs?
Coherence Thresholds - How different is TOO different?
Periodic Synchronization - "Check in, are we still on the same track?"
Guided Divergence - "Here's a direction we haven't explored yet"

Implementation

See:

.claude/skills/divergence-control/divergence_manager.py

The Balance

0% divergence = Waste of resources
100% divergence = Incoherent output
30-50% divergence = Optimal exploration

Payment Anchor

DOGE: DC8HBTfn7Ym3UxB2YSsXjuLxTi8HvogwkV