Harness Template — Development skill for Claude Code
A harness for coding agents, installed in one command: goal, backlog, boot sequence, checks, guard hooks, reviewer and strategist subagents.
How to install Harness Template
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open theolegr/harness-template and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Harness Template does
A harness for coding agents, installed in one command: goal, backlog, boot sequence, checks, guard hooks, reviewer and strategist subagents. Claude interviews you to fill it in; /ship builds one item at a time with an independent review; --update brings in new versions with a 3-way merge. Claude Code first; AGENTS.md for Codex, Cursor.
Alternatives in Development
- Dashclaw Quality — Recurring find-and-fix quality pass over the whole DashClaw app — browser smoke (frontend-verify) + code gates 297 ★
- Maestro — BOOT SEQUENCE: Activate The Grandmaster (System Orchestrator) with strategic mode selection 210 ★
- Springboot Skills Marketplace — Spring Boot Skill for Claude Code and Codex 72 ★
README
Project Harness — bootstrap template
A portable, agent-agnostic harness that gives any new project (SaaS, app, website, API, bot) a strong base: identity, a measurable goal, a boot sequence, a feature tracker, verification instruments, memory architecture, a self-improvement loop, and an outer loop with orchestration patterns.
Works with Claude Code, Codex, Cursor — anything that reads `AGENTS.md`. The hooks, skills and subagents are Claude Code's; other agents get the same rules as prose in `AGENTS.md`, plus the checks inside `loop.sh`.
It comes out of my own setup: the harness I use to run my projects with agents, distilled from what I read and tested. It's opinionated — take it, then adapt it to your project (and delete what doesn't earn its place).
**New here? Read [GUIDE.md](GUIDE.md)** — the whole workflow on a concrete example, driven from a chat with Claude Code.
**Status: early.** The scripts (install, backlog, checks, hooks, `loop.sh`) are tested on throwaway projects. The chat workflow — `/ship`, the reviewer and strategist subagents, the hooks — has run on three real projects (about 70 items shipped). Not yet run for real: the `design` skill as written and `loop.sh` with a real Claude. Details in [TODO.md](TODO.md). Issues and feedback welcome.
How it works
- Install —
init-harness.sh my-projectcopies the harness into a new or existing project, then Claude interviews you (~10 min) and fills in the goal, the stack and its commands, the first features and the autonomy rules. - Work — every session starts from the real state (a hook shows goal,
current item, git state). In the chat,
/shiptakes one backlog item at a time: plan → your OK → build → a read-onlyreviewersubagent checks it → the item is recorded and committed. Astrategistsubagent helps decide what comes next. Unattended,loop.shruns the same loop. - Guard rails — Claude Code hooks refuse destructive commands, ask before m
Related Skills
Boot
Session boot sequence. Run at the start of every session to load memory and verify rules.
Dawn
Morning briefing — composes /arcanea-status + /pulse + /sis recent + last handover's recommended next stack in
Roadmap Pro
Shape evidence-linked backlog choices, advise on priorities, reconcile human selection, explain every reviewed
Add Backlog Item
Backlog stage. Creates one backlog work item from natural-language input with tracker destination detection an
Proceed With The Recommendation
Execute the agent's prior recommendation list under the 7 Laws — walk in order, route per item, verify, reflec
Gsd T Backlog Add
GSD-T: Backlog Add — Capture and Categorize a Backlog Item
Related Agents
YouTube Archify Diagrammer
Generates real-world architecture, workflow, sequence, dataflow, or lifecycle diagrams for a "Claude Explains"
Diagram Builder
Generates ship-time Mermaid state and sequence diagrams, an API endpoint catalog with request/response schemas
Harness Eng
Use for every build task on a deepseek-harness core — one layer per claimed packet, gated by hedgehog verify.