create-harness

Scaffolds a Claude Code agent harness monorepo with skills, MCP servers, and graphify knowledge graphs.

Updated Apr 11, 2026
One-click install
npx skills add https://github.com/lurodrisilva/personal-skills --skill create-harness-lurodrisilva
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: create-harness
Source: https://github.com/lurodrisilva/personal-skills/tree/main/platform-engineering/create-harness
Command: npx skills add https://github.com/lurodrisilva/personal-skills --skill create-harness-lurodrisilva

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Teams managing Claude Code configuration across multiple repositories end up with scattered, hand-edited .claude/ directories and duplicated skills. This Skill bootstraps a single canonical monorepo (a "harness") that holds skills, MCP servers, sub-agents, and a knowledge vault, then projects everything into Claude Code via a generated bin/harness sync CLI. ## Core Features & Use Cases - Interview-first scaffolding: Collects the harness location, contemplated repositories, MCP servers, plugins, and vault preferences before writing any file, then generates the full topology (harness.config.yaml, bin/harness, bin/render_mcp.py, bin/render_plugins.py, mcp/servers.json, workspace/workspace.yaml, vault/). - Canonical-to-projection sync: Generates CLAUDE.md as a symlink to AGENTS.md, symlinks active skills and sub-agents into .claude/, and renders .mcp.json and .claude/settings.json from canonical sources. - Graphify knowledge graphs: Clones the contemplated repos into workspace/, builds per-repo graphify graphs, and merges them into a federated graphify-out/federated-graph.json for cross-repo architectural queries. - Use Case: A platform engineer asks to "create a harness" for a team with five product repos; the Skill interviews for the repo list, scaffolds the monorepo, clones the repos, runs harness sync and harness graph, and verifies the federated graph answers queries. ## Quick Start Ask the agent to create a new Claude Code harness at a chosen path, providing the list of repositories to contemplate, and let it interview you for MCP servers, plugins, and vault options before scaffolding.

Frequently Asked Questions about create-harness

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

FAQPage Schema
How do I create a Claude Code agent harness from scratch?▼

Run this Skill and answer its Step 0 interview: harness path and name, the list of contemplated repositories with clone URLs, which MCP servers to enable, and vault preferences. It then scaffolds the topology, clones the repos, runs bin/harness sync, and builds the graphify graphs.

How to manage Claude Code skills and MCP servers across multiple repositories?▼

Keep one canonical monorepo with catalog/skills, mcp/servers.json, and harness.config.yaml, then run ./bin/harness sync to project them into .claude/ and .mcp.json. Never hand-edit the projection; edit the canonical source and re-sync.

Does this harness support Devin or other AI coding CLIs?▼

No, this Skill is Claude-Code-only by design and deliberately omits Devin and other CLI adapters. For a multi-CLI harness you would need to adapt the topology with a separate adapter and a branch in bin/harness sync.

Why does harness graph fail after scaffolding?▼

The graph command requires at least two graphs, meaning at least one cloned repo plus the harness itself. Confirm the contemplated repos were cloned into workspace/ in Step 2 and that graphify is installed via uv tool install graphifyy.

When should I not use the create-harness skill?▼

Skip it when the harness already exists and you only need to add a repo (use ./bin/harness new-project then ./bin/harness graph) or only regenerate the projection (use ./bin/harness sync). It is for bootstrapping a brand-new harness.