start-project

Initialize new projects and assemble portable agent harnesses from confirmed manifests.

Updated Aug 19, 2026
One-click install
npx skills add https://github.com/PVMalove/claude-agent-harness --skill start-project-pvmalove
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: start-project
Source: https://github.com/PVMalove/claude-agent-harness/tree/main/global-skills/start-project
Command: npx skills add https://github.com/PVMalove/claude-agent-harness --skill start-project-pvmalove

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Starting a new project involves scattered decisions about repositories, stacks, trackers, and agent tooling, and it is easy to scaffold files before the owner has confirmed what should exist. This Skill guides the path from an unshaped idea to a portable project with a self-contained agent harness, keeping every phase read-only until the owner confirms each durable artifact. ## Core Features & Use Cases - Phased project design: Shape (classify the idea and resolve decisions), Seed (create the Git repository with only confirmed facts), Assemble (build the harness from a confirmed manifest), and Finish (verify stored, wired, healthy, advertised, and invoked layers). - Harness assembly: Selects the smallest coherent public capability such as project-foundation or mattpocock-suite, runs the harness CLI init command, locks project-only skill packages, and inventories integrations in .harness/integrations.json without copying credentials into Git. - Use Case: A user has a rough idea for a content research project. The Skill classifies the domain, recommends creating a seed repository, then assembles a project-foundation harness and verifies it with harness health and a fresh-session discovery check. ## Quick Start Use the start-project skill to shape my project idea and initialize the right agent harness once I confirm the plan.

Frequently Asked Questions about start-project

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

FAQPage Schema
How do I start a new project with an agent harness?▼

Follow the phased workflow: shape the idea by classifying the project domain, seed a Git repository with only confirmed facts, then assemble the harness from a confirmed manifest. The harness init command installs the selected capability into the project.

What is the difference between project-foundation and mattpocock-suite capabilities?▼

project-foundation covers general project thinking, research, handoff, and agent-facing instructions for any domain. mattpocock-suite targets recurring software engineering work with 25 engineering skills. The Skill selects the smallest coherent capability for the project.

When should I use integrate-project instead of start-project?▼

Use integrate-project when a repository already has substantial existing code and conventions but no harness yet, since it performs a deeper dedicated audit. start-project handles unshaped ideas, seed directories, and new repositories.

Does start-project work on Windows with PowerShell?▼

Yes, but the harness command differs: use python instead of python3, and use the backtick character for line continuation instead of a backslash. The Skill documents both variants for the harness init command.

How does start-project handle secrets and MCP server credentials?▼

It never copies credentials into Git. MCP servers, plugins, hooks, and runtime settings stay in their native project files, and only their relative paths, hashes, runtimes, and secret environment-variable names are inventoried in .harness/integrations.json.

How do I verify a harness was installed correctly?▼

Run harness health to check template markers, names, registry, hashes, and discovery links. Then confirm a fresh runtime session lists an installed project skill and can invoke it through its native mechanism.