Topcode — AI skill for Claude Code
topcode is a The next-generation CLI agent framework — model-agnostic, context-entropy-resistant, built on State Manifold Projection.
How to install Topcode
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open chees-cn/topcode and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Topcode does
topcode is a The next-generation CLI agent framework — model-agnostic, context-entropy-resistant, built on State Manifold Projection. Underlying paradigm: Topological Wavefront Execution · Positioning: a direct competitor to Claude Code.
Alternatives in AI
- Segmental Model — 分业务收入 bottom-up 建模 (v2.10) — 拆产品线 × 3 情景 × 3 年 projection,对 DCF 自上而下做反向校验 6.6k ★
- Osaurus — Own your AI. The native macOS harness for AI agents -- any model, persistent memory, autonomous execution, cry 5.1k ★
- Complete Context For Vertex AI: Life Sciences MCP Plugin Generation — COMPLETE Context for Vertex AI: Life Sciences MCP Plugin Generation skill 1.7k ★
README
TopCode
[中文](README.zh.md) | **English**
The next-generation CLI agent framework — model-agnostic, context-entropy-resistant, built on **State Manifold Projection**.
Underlying paradigm: **Topological Wavefront Execution** · Positioning: a direct competitor to Claude Code.
**Current version: v0.7.0** · License: AGPL-3.0 · Runtime: Node.js + NestJS + TypeScript
Table of Contents
- Philosophy: Why We Overthrow the Old Paradigm
- The Six Paradigm Pillars
- Core Features
- System Architecture
- Quick Start
- CLI Usage
- Action Routing Protocol
- Configuration
- State Files & Memory System
- Directory Structure
- Evaluation System
- Roadmap & Milestones
- Documentation Map
- Development Rules
Philosophy: Why We Overthrow the Old Paradigm
Every existing CLI agent shares one old axiom:
"An agent's state = a one-dimensional tape of conversation tokens."
This produces four inevitable pathologies:
- Irreversible context entropy — compression means loss, loss means amnesia, amnesia means repeating mistakes;
- One-dimensional time — a serial while-loop; trial-and-error cost = wall-clock time × serial iterations;
- No world model — knowledge has no confidence structure; every re-observation is paid for again;
- No intelligence accumulation — the hundredth identical operation burns the same tokens as the first.
TopCode's new axiom:
**An agent's state is an evolvable, verifiable, projectable State Manifold. The LLM is not the container of state — it is a state-transition operator on the manifold. Context is not a log; it is a view computed on demand from the manifold.**
Full argumentati
Related Skills
Nissth
Deterministic execution framework for LLM coding agents. Pre-structured context, append-only state, no explora
Godot AI
Native C++ AI assistant module for the Godot 4.7 editor - multi-provider LLM chat, GDScript code generation &
DCGP
Context governance kernel for LLM agents. Predicts entropy, blocks hallucinations. Pairs with opencode-dcp. ll
Hand Compute
Skill to force AI coding agents to walk system execution by hand, catching races and state bugs that abstract
CCW Coordinator Command
Interactive orchestration tool: analyze task → discover commands → recommend chain → execute sequentially → tr
Devloop
A guard-railed, closed-loop workflow for AI coding agents: live state bus + execution-level hard intercepts fo
Related Agents
Pi Implementer
Implementation lane driven by the pi CLI (a local multi-provider coding agent, distinct from omp). Model-agnos
Opencode Generalist
OpenCode, the open-source model-agnostic CLI harness, for general coding
Aps Conductor
Coordinate APS execution through CLI-backed next-work selection, context packaging, dependency checks, validat