Skip to content

dspury/agent-harness-template

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

8 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Agent Harness Template

A self-optimizing, agent-agnostic harness for structured AI-assisted development.

Use the Quick Start prompt to have your coding agent adapt this template to the target repository's real structure, tooling, and workflows.

It keeps planning, execution, and validation artifacts inside the repository as file-based state: no chat-history dependency, no external services, and no orchestration layer. Works with Claude, Codex, or any LLM-based coding agent.

Purpose

Most agent failures come from weak repo context and inconsistent execution structure. This template fixes that by bootstrapping a repeatable task loop, then having the agent self-optimize the harness for each target repo.

A brief defines what, a plan defines how, and a run record proves it worked.

Quick Start

  1. Copy the harness files into your repo root: HARNESS.md, ARCHITECTURE.md, SETUP.md, .harness/, and docs/
  2. Open SETUP.md and paste the prompt into your coding agent
  3. The agent self-optimizes the harness to your repo's real structure and workflows

That's it. After setup, follow the Task Loop in HARNESS.md.

What's Inside

Path Purpose
HARNESS.md Agent entry point — repo map, working rules, task loop
ARCHITECTURE.md Structure, data flow, and constraints
SETUP.md Easy Mode bootstrap prompt (deleted after setup)
docs/briefs/ Task brief template
docs/plans/ Implementation plan template
.harness/tasks/ Task packet template (JSON)
.harness/runs/ Run record template (JSON)

Manual Setup

If you prefer to configure the harness by hand instead of using Easy Mode:

  1. Copy the harness files listed above into your repo root
  2. Read HARNESS.md and ARCHITECTURE.md
  3. Edit HARNESS.md to map your real repo (replace all TODO lines)
  4. Edit ARCHITECTURE.md to describe your actual structure
  5. Create or append to your agent's config file (CLAUDE.md, AGENTS.md) with a pointer to HARNESS.md
  6. Delete SETUP.md
  7. Start using the Task Loop

Design Principles

  • File-based state — everything lives in the repo, versioned with git
  • Agent-agnostic — no agent-specific files; works with any LLM coding tool
  • Self-optimizing setup — Quick Start prompt adapts templates to the target repo
  • Minimal surface — small set of templates, zero dependencies
  • Starting point, not framework — adapt the templates to fit how you work

License

MIT

About

Self-optimizing, agent-agnostic repo harness that adapts to any codebase via a quick-start setup prompt.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors