Can this be used across different editors?

Yes, it supports bisimulation equivalence, which allows skills and agent states to be synchronized and verified across Claude, Codex, Cursor, and Copilot using a shared fingerprint.

What are Narya observational bridge types?

They are a category-theory-inspired approach to version control that treats code diffs as logical relations and modifications as commuting squares, ensuring mathematical integrity during self-modification.

How does the MCP Task integration work?

It maps modification states like 'working' or 'input_required' to a 'TAP' state system (LIVE, VERIFY, BACKFILL), providing a structured lifecycle and safety checks for every code change the agent makes to itself.

What is self-rewriting in the context of this skill?

It allows the AI agent to modify its own source code and internal logic using structured patterns similar to a Lisp machine, ensuring the agent can evolve its capabilities autonomously while maintaining state.

Codex Self-Rewriting

Name: Codex Self-Rewriting
Author: plurigrid

byplurigrid

•

開発者ツール

Enables autonomous self-modification for AI agents using Lisp-machine patterns and MCP Tasks.

This skill empowers Codex and other AI agents to achieve Lisp-machine-like self-rewriting capabilities through structured Model Context Protocol (MCP) Tasks and Narya observational bridge types. By utilizing Babashka transients and topological version control, it allows agents to safely modify their own logic, maintain cognitive continuity across sessions, and manage state transitions using a color-coded verification system. It is specifically designed for developers building highly autonomous systems that need to evolve their own capabilities while maintaining rigorous, structure-aware versioning based on category theory principles.

主な機能

01Cognitive continuity through XOR-based skill fingerprinting

02Narya observational bridges for structure-aware diffs and conflicts

037 GitHub stars

04Lisp-machine style self-modification via Babashka transients

05MCP Task integration for structured agent state management

06Resilient skill dispersal across multiple editors using bisimulation

ユースケース

01Maintaining consistent agent behavior across different IDEs like Cursor and Copilot

02Implementing formal, category-theory-based version control for AI-generated code

03Building autonomous agents that can safely update their own skill definitions

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add plurigrid/asi codex-self-rewriting

For use in Claude.ai and ChatGPT

Download Skill