explorer

Generate a single disciplined candidate branch with an EVOLVE_STATE block.

Updated Apr 15, 2026

Search Tags:#automation #exploration #evolution #scope #branch #hypothesis #code-review

Ready to run

npx skills add https://github.com/kyleyhw/agent-evolve --skill explorer-kyleyhw

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: explorer
Source: https://github.com/kyleyhw/agent-evolve/tree/main/.claude/skills/explorer
Command: npx skills add https://github.com/kyleyhw/agent-evolve --skill explorer-kyleyhw

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

It provides a disciplined protocol for producing a single branch candidate in an evolutionary run, ensuring the exploration respects scope and yields a trackable EVOLVE_STATE.

Core Features & Use Cases

Read the current parent candidate(s), the assigned operator, and the problem spec to prepare a focused change.
Write the hypothesis before coding and enforce changes strictly within the target files.
Commit a single candidate branch with a completed EVOLVE_STATE block for reviewer tooling.

Quick Start

Write the hypothesis first and then implement the change inside the allowed files and push the candidate branch.

Frequently Asked Questions about explorer

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

FAQPage Schema

How do I generate a single candidate branch in an evolutionary code run?▼

You generate a single candidate branch by reading the current parent candidate, the assigned operator, and the problem spec to prepare a focused diff. This ensures the exploration respects scope and yields a trackable EVOLVE_STATE for reviewer tooling.

What is hypothesis-first code evolution and how does it enforce scope?▼

Hypothesis-first code evolution requires you to write the hypothesis before coding and enforce changes strictly within allowed target files. This disciplined protocol ensures exploration respects scope and produces focused diffs within evolve/<problem-id> branches.

How do I track evolution state across multiple code generation rounds?▼

You track evolution state by committing a single candidate branch with a completed EVOLVE_STATE block. This block coordinates with parents, operators, and problem specs to provide trackable commits for reviewer tooling.

Can I sanity-check a candidate branch locally before submitting it to an evolution run?▼

Yes, you can sanity-check a candidate branch locally before submission. The evolutionary run produces focused diffs within evolve/<problem-id> branches that can be validated locally to ensure changes touch only allowed target files.

What is the best way to coordinate parent candidates and operators during code evolution?▼

The best way to coordinate parent candidates and operators is to read the current parent candidate, the assigned operator, and the problem spec together to prepare a focused change. This enforces that changes touch only allowed target files and produces a trackable EVOLVE_STATE.