Speculative Patterns

[adamrdrew]

Speculative Patterns: Exploring Claude Code as a Platform

This PR adds 10 experimental design patterns that push beyond the established foundation patterns. These emerged from a structured exploration process:

1. Analyzed the existing patterns to extract underlying principles
2. Studied Claude Code's architecture (hooks, skills, agents, permissions)
3. Asked: what other theories about AI system design could be implemented?

The Patterns

Pattern	Core Idea
Dialectic Agents	Opposing agents (Advocate/Critic/Arbiter) debate to surface assumptions
Reflexive Monitoring	Agents watch themselves via hooks; self-doubt as a feature
Evolutionary Skills	Skills compete, fitness is evaluated, winners get promoted
Constraint Inheritance	Governance flows from parent to child agents
Session Archaeology	Cross-session learning through distilled experience
Capability Synthesis	Observe skill co-occurrence, propose composite skills
Gradual Trust	Earn permissions through demonstrated competence
Deliberate Forgetting	Prune stale knowledge to stay lean
Adversarial Testing	Red/Blue teams probe and harden each other
Ceremony Patterns	Deliberate friction for high-stakes actions

Status: Experimental

These patterns are not yet validated in real projects. They represent promising ideas that need testing. Each pattern includes working agents, skills, and supporting infrastructure—but whether they deliver value in practice remains to be seen.

Repo Changes

• New speculative-patterns/ directory containing all 10 patterns
• Each pattern has full implementation (agents, skills, settings, state directories)
• Comprehensive README explaining each pattern and how they were developed
• Main README updated to reference speculative patterns without duplicating details

Philosophy

These patterns explore different theories about how AI systems should operate: structured disagreement, self-reflection, natural selection, progressive trust, and deliberate friction. They treat Claude Code not just as a coding assistant but as a platform for implementing cognitive architectures.