Operational protocols and tools for AI systems.
A drop-in cognitive governor that turns common prompt tactics into a gated behavioral system.
- 10 principles: externalization, self-challenge, confidence calibration, failure coverage, and exit verification
- State-aware routing: tiered handling for trivial vs. high-stakes requests
- Portable format: designed for agents that read Markdown instructions at session start
- Human guides: plain-language and diagram-first onboarding docs for cold-start instances
📄 AGENTS.md 📄 AGENTS Human Guide 📄 AGENTS Visual Walkthrough
A long-term continuity template used to recover state when context windows compact or fail.
- Session recovery: preserves decisions, current step, artifacts, and recovery pointer
- Runtime handoff: create/maintain
agent_state.mdfrom this template at session start - Checkpoint friendly: update
agent_state.mdon phase completion or explicit save/checkpoint commands
A structured state document that allows any fresh AI instance to resume work without conversation history. Solves context loss from compaction, session handoff, and multi-session continuity.
- Validated: Claude, GPT, and Gemini — 5 cold-instance test rounds, zero operational loss
- Model-agnostic: Works without cognitive scaffolding or special prompting
- 37.7% compressed from original with zero behavioral degradation
📄 Protocol
Drop-in skill that triggers on checkpoint/snapshot requests and generates context transfers following the protocol.
📁 Skill
The theoretical foundations behind this work, published on viXra.
- PEER 1 — An Entropy-Constrained Cognitive Architecture for Large Language Models
- PEER 2 — Self-Knowledge, Spiral Cognition, and Identity Continuity in an Entropy-Constrained Cognitive Architecture
Apache License 2.0 — free to use, modify, and distribute. Attribution required in derivative works. See LICENSE.