harness-config

Delegate non-markdown Claude Code harness changes to an infrastructure-engineer with worktree isolation.

1|Updated Mar 15, 2026
One-click install
npx skills add https://github.com/paulingham/.claude --skill harness-config
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: harness-config
Source: https://github.com/paulingham/.claude/tree/main/skills/harness-config
Command: npx skills add https://github.com/paulingham/.claude --skill harness-config

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Modify the Claude Code harness itself: hooks (.sh), settings.json, agent definitions, and skill infrastructure. Delegates non-markdown file changes to an infrastructure-engineer agent with worktree isolation. Use when adding/modifying hooks, settings, or any non-markdown config in ~/.claude/.

Core Features & Use Cases

  • Manages changes to the Claude Code harness — the hooks, settings.json, agent definitions, and skill infrastructure that control how the orchestrator and agents operate.
  • Delegates non-md file changes to an infrastructure-engineer (worktree isolation) to preserve safety and reproducibility.
  • Applies when adding or modifying hook scripts, JSON/YAML settings, agent definitions containing non-trivial logic, or any change to ~/.claude/ that is not a Markdown document.
  • Ensures changes are described, validated, and registered in settings for governance.

Quick Start

Describe the change, spawn an infrastructure-engineer with a worktree isolation, verify the change, and register it in settings.

Frequently Asked Questions about harness-config

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

FAQPage Schema
How do I safely modify Claude Code harness settings and hooks?

To safely modify Claude Code harness settings, the configuration process delegates non-markdown file changes to an infrastructure-engineer agent using worktree isolation, ensuring changes are described, verified, and registered in settings.json.

What is worktree isolation and when do I need it for ~/.claude/ configuration?

Worktree isolation is required when adding or modifying non-markdown files under ~/.claude/, such as hook scripts (.sh), settings.json, or agent definitions, preventing unverified changes from directly affecting the active orchestrator.

Can I directly edit settings.json and agent definition files in my Claude Code environment?

Directly editing settings.json or agent definition files containing non-trivial logic is discouraged; these non-markdown changes must be delegated to an infrastructure-engineer prompt and registered in settings for governance.

What's the best way to add hook scripts to the Claude Code harness?

The best way to add hook scripts (.sh) is to describe the change, spawn an infrastructure-engineer with worktree isolation to execute it, verify the modification, and register it in settings to preserve safety and reproducibility.

Does this configuration method apply to markdown documents inside ~/.claude/?

No, this configuration method specifically applies to non-markdown changes under ~/.claude/, such as hooks, settings.json, and agent definitions, while markdown documents fall outside this delegation scope.