zyron-reasoning
v1.0.3
Published
A cognitive governance layer for AI coding assistants. Changes how the model thinks before generating — not just what it does.
Maintainers
Readme
zyron-reasoning
A cognitive governance layer for AI coding assistants. It changes how the model thinks before generating — not just what it does.
What this is
A set of structured markdown files that your AI assistant reads before every generation. They enforce a thinking protocol: define the real problem, find the minimal solution, assess blast radius, check existing solutions, and set done criteria — before writing a single line of code.
Works with Claude Code, Cursor, Windsurf, and Antigravity.
What this is NOT
- Not a prompt library or skills collection.
- Not a chatbot wrapper or agent framework.
- Not a set of code snippets or templates.
- Not IDE-specific — the core files are universal.
Install
npx zyron-reasoningThe CLI detects your IDE and copies the reasoning files to the right paths.
Use --all to install for all supported IDEs at once.
How it works
The system uses 5 core files, each a markdown layer the model reads at different moments:
| Layer | File | Purpose |
|-------|------|---------|
| 0.5 | personas.md | Your experience level and domain. Adapts tone and depth. |
| 1 | context.md | Your project's stack, patterns, and constraints. |
| 2 | thinking.md | 5-gate preflight protocol. The heart of the system. |
| 3 | review.md | Quality checklist applied to output before delivery. |
| 4 | decisions.md | Running log of technical decisions. Prevents repeated mistakes. |
The model reads them in order: personas → context → decisions → thinking → review.
Quick start
Run the installer
npx zyron-reasoningOpen
.zyron/START_HERE.md— guided setup in 2 minutes.Set your persona — open
.zyron/personas.md, uncomment your level:level: beginner | intermediate | professionalDescribe your project — fill in
.zyron/context.mdwith your stack and rules.
That's it. The reasoning layers activate automatically on every generation.
The thinking protocol
Every time your assistant generates code, it runs through 5 gates:
- Problem Definition — What is the REAL problem, not the symptom?
- Minimal Solution — What is the SMALLEST change that solves it?
- Blast Radius — What can this break?
- Existing Solutions — Does a built-in or installed solution already exist?
- Done Criteria — How do I know this is done?
No code is written until all 5 gates are answered.
Supported IDEs
| IDE | Rules path | Adapter |
|-----|-----------|---------|
| Claude Code | .claude/ | zyron-reasoning.md |
| Cursor | .cursor/rules/ | zyron-reasoning.mdc |
| Windsurf | .windsurf/rules/ | zyron-reasoning.md |
| Antigravity | .agent/ | zyron-reasoning.md |
Persona levels
| Level | Behavior | |-------|----------| | beginner | Patient, step-by-step, warns before risky actions, explains the why. | | intermediate | Direct, technical, explains only non-obvious parts. Default. | | professional | Terse, fast, no hand-holding, optimized for speed. |
Contributing
See CONTRIBUTING.md for guidelines on proposing new layers, IDE adapters, or improvements.
License
MIT — Zyron AI
