kernel-extension

Extend the composable kernel with phases, guards, and MetaTools.

21|4|Updated Feb 19, 2026
One-click install
npx skills add https://github.com/tylerjrbuell/reactive-agents-ts --skill kernel-extension
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: kernel-extension
Source: https://github.com/tylerjrbuell/reactive-agents-ts/tree/main/.agents/skills/kernel-extension
Command: npx skills add https://github.com/tylerjrbuell/reactive-agents-ts --skill kernel-extension

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill provides extension points for the composable kernel so developers can implement custom per-turn logic, intercept or block tool calls, synthesize tool outputs, or replace the phase pipeline for new reasoning strategies.

Core Features & Use Cases

  • Phase Extension: Define pure, Effect-based phases that read kernel state and return updated KernelState for enrichment, analysis, or reflection.
  • Guard Rules: Implement synchronous guards that explicitly allow or block tool calls with reasons to enforce execution policies and safety constraints.
  • MetaTools & Custom Kernels: Intercept named tool calls to synthesize results in-memory or compose a fully custom phase sequence for alternative reasoning strategies.
  • Use Case: Create a strategy that inserts a pre-think enrichment phase, adds a guard to block unsafe tool names, and registers a MetaTool to return cached responses without external I/O.

Quick Start

Use the kernel-extension skill to add a pre-think phase that enriches per-turn context and register it in your kernel's phases array.

Frequently Asked Questions about kernel-extension

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

FAQPage Schema
How do I add custom reasoning phases to a TypeScript agent framework?

To add custom reasoning phases to a TypeScript agent framework, define pure Effect-based phase functions that read kernel state and return updated KernelState, then register them in your kernel configuration's phases array.

Can I block unsafe tool calls in an LLM reasoning pipeline?

Yes, you can block unsafe tool calls in an LLM reasoning pipeline by implementing synchronous guard rules that return explicit GuardOutcome shapes to allow or block execution with reasons.

What is the best way to synthesize tool responses without external I/O in an agent?

The best way to synthesize tool responses without external I/O in an agent is to register MetaTools that intercept named tool calls and return cached or in-memory synthetic results directly within the kernel.

Does the composable kernel require Effect-based functions for phase extensions?

Yes, the composable kernel requires pure Effect-based phase functions for extensions, while guards must use synchronous return types matching exact GuardOutcome shapes to ensure predictable execution.

How do I replace the phase pipeline for alternative reasoning strategies?

To replace the phase pipeline for alternative reasoning strategies, compose a fully custom kernel variant by defining a new phase sequence that customizes per-turn state transformation and tool call filtering.