What problem does it solve?
Scaffolding a new AI agent harness involves write operations that can accidentally overwrite or pollute the current repository. This Skill wraps the metaharness new command with mandatory dry-run behavior and path safety checks so nothing is written to disk until explicitly confirmed.
Core Features & Use Cases
- Dry-run by default: Without
--confirm, the script prints the full scaffold plan (name, template, host, target) and exits without touching disk.
- Path safety enforcement: Refuses to write to the calling repository root or any path inside it, and refuses to overwrite existing directories.
- Template and host selection: Supports 20 templates (minimal, vertical:coding, vertical:devops, vertical:legal, etc.) and 9 hosts (claude-code, codex, copilot, github-actions, etc.).
- Use Case: You want to prototype a coding-focused agent harness for Claude Code. Run the skill with
--name my-harness --template vertical:coding --host claude-code, review the dry-run plan, then re-run with --confirm to scaffold into a fresh temp directory.
Quick Start
Ask the AI to scaffold a new agent harness named my-harness using the vertical:coding template for claude-code, review the dry-run output, then confirm to write it.
