zfc

Delegate cognitive decisions to the model while enforcing deterministic shell IO.

1|Updated Apr 28, 2026

Search Tags:#guardrails #determinism #prompting #zfc #ai-shell #model-delegation #io-management

Ready to run

npx skills add https://github.com/naramore/dotai --skill zfc-naramore

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: zfc
Source: https://github.com/naramore/dotai/tree/main/claude/skills/zfc
Command: npx skills add https://github.com/naramore/dotai --skill zfc-naramore

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Delegate cognitive decision-making to the model while enforcing deterministic IO in the shell. The shell handles IO, schemas, budgets, and retries, ensuring a predictable execution path while the model performs reasoning.

Core Features & Use Cases

Thin, deterministic shell architecture that isolates cognition in the model.
Policy enforcement, input/output validation, and mechanical transforms to keep execution safe and auditable.
Use cases include AI applications, agent loops, tool-use harnesses, and code that consumes LLM output, where consistent behavior and strong guardrails are required.

Quick Start

Launch the ZFC-enabled environment to wrap your agent loop so all judgment is delegated to the model, while the host shell manages IO and safety.

Frequently Asked Questions about zfc

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I enforce deterministic IO in an AI agent loop while delegating reasoning to the model?▼

To enforce deterministic IO in an agent loop, use a thin shell architecture that isolates cognition in the model. The shell strictly manages schemas, budgets, and retries, ensuring predictable execution paths while the model handles reasoning.

What is the best way to apply guardrails to LLM output for code that consumes model responses?▼

The best way to apply guardrails to LLM output is to wrap execution in a deterministic host shell. This enforces input/output validation, policy enforcement, and mechanical transforms, ensuring safe and auditable behavior for consuming applications.

How do you handle retries and output schemas for tool-use harnesses without restricting model cognition?▼

Handle retries and output schemas for tool-use harnesses by delegating cognitive decisions to the model while the surrounding shell mechanically enforces strict output schemas, manages budgets, and executes safe retries.

Does a deterministic shell architecture work for AI applications requiring strict structural safety?▼

Yes, a deterministic shell architecture works for AI applications requiring structural safety. It isolates judgment inside the model while the shell manages mechanical data handling, providing the strong guardrails and predictable behavior needed.

When should I not use a thin deterministic shell for model delegation?▼

You should avoid a thin deterministic shell for model delegation if your application does not require strict output schemas, predictable execution paths, or enforced budgets. It is designed for scenarios demanding structural safety and auditable behavior.