pi-harness

Name: pi-harness
Author: wangmiaozero

Pending

Visual desktop Harness & control center for Pi Coding Agent — inspect runtime, context, tools & sessions; manage models & skills; browse files & Git. Powered by Pi Agent Harness.

51stars

4forks

TypeScript

Installation

# Add to your Claude Code skills
git clone https://github.com/wangmiaozero/pi-harness

Getting Started

Guides for using ai agents skills like pi-harness.

Caveman: Cut Claude Token Use by 65%
How agent-side prompt compression works, when to use it, and when not to.
What is an AI Skills Marketplace?
Definitions, how marketplaces work, and how to choose between them in 2026.
Getting Started with AI Skills

README.md

Frequently Asked Questions

What is pi-harness?

pi-harness is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by wangmiaozero. Visual desktop Harness & control center for Pi Coding Agent — inspect runtime, context, tools & sessions; manage models & skills; browse files & Git. Powered by Pi Agent Harness. It has 51 GitHub stars.

Is pi-harness safe to use?

pi-harness's catalog security scan is still queued. You can run an instant dependency and prompt-injection check now with the "Scan for vulnerabilities" button above.

How do I install pi-harness?

Clone the repository with "git clone https://github.com/wangmiaozero/pi-harness" and add it to your Claude Code skills directory (see the Installation section above).

What programming language is pi-harness written in?

pi-harness is primarily written in TypeScript. It is open-source under wangmiaozero on GitHub, so you can review or fork the full source.

Are there alternatives to pi-harness?

Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh pi-harness against similar tools.

Agentic AI for Beginners

Build your first AI agent from scratch - tool use, ReAct pattern, memory, deployment

41 minBeginner

Comments (0)

to leave a comment.

No comments yet. Be the first to share your thoughts!

Related Skills

ECC

by affaan-m

The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

251,960

github-mcp-server refine-canvas-strokes

Pi-Harness

Why Pi-Harness?

Pi-Harness is a visual Harness Console for Pi Coding Agent. Use Native Pi as-is, or discover and manage optional development methodologies, Skills, extensions, packages, and MCP tools.

Pi Coding Agent already has a powerful Agent Harness. Much of its runtime state—context, tools, compaction, queues, and sessions—is naturally exposed through CLI and SDK behavior.

Pi-Harness makes those capabilities visible and manageable in a native desktop workspace, alongside the configuration and project tools needed to use Pi every day.

Pi runs the agent. Pi-Harness lets you see and control how it runs.

Pi-Harness is not a wrapped web UI. It embeds no pi-web, Next.js server, or iframe, and it does not add a second Agent Runtime.

Powered by Pi Agent Harness

Pi already provides the Agent Harness that powers Pi Coding Agent.

Pi-Harness does not replace or reimplement that runtime. Pi Coding Agent and its Agent Harness remain responsible for the Agent Loop, tools, context, compaction, and session execution.

Pi-Harness provides the desktop control plane, visualization, configuration, and workspace around it:

inspect runtime and context state
start, resume, fork, and navigate sessions
select models, Thinking levels, and tools
control compaction, Steering, and Follow-up messages
observe streaming output, Thinking, Tool Calls, and agent events
work with files and Git beside the agent

Powered by Pi Agent Harness.

What Pi-Harness already does

These are current capabilities, not roadmap claims: Pi-Harness can start and resume Pi-compatible sessions; stream responses, Thinking, and Tool Calls; show context usage and session statistics; switch models, Thinking levels, and tool presets; control compaction; send Steering and Follow-up messages; and Fork or navigate session history.

The surrounding desktop capabilities are summarized below.

Core features

Area	What Pi-Harness does
Overview	Shows environment, configuration, and current model status
Workspace	Runs Pi sessions beside project files, Git, and worktrees
Pi Runtime	Surfaces streaming, Thinking, Tool Calls, context, queues, and stats
Providers & Models	Manages Pi-compatible providers and models
Skills & Packages	Manages supported Skills, Pi packages, and MCP-capable extensions
Files	Provides lightweight editing with explicit save and conflict handling
Diagnostics	Reports application, environment, storage, and workspace health
Updates	Installs compatible application updates
Appearance	Offers appearance, density, cockpit mode, and optional themes

Optional methodologies and add-ons

Native Pi remains the built-in default runtime and workflow. Pi-Harness does not replace it and does not silently install or enable third-party methodologies.

Superpowers is the recommended optional add-on: a complete software development methodology for coding agents covering brainstorming, planning, TDD, systematic debugging, Git worktrees, subagent and parallel-agent workflows, code review, and verification. Install it explicitly from Capabilities with pi install git:github.com/obra/superpowers.
Odai remains an optional governance and adaptive-execution methodology for goal alignment, authorization boundaries, risk awareness, capability routing, evidence, and verification.

Both add-ons use the existing trusted capability/package management flows and can be updated or removed independently.

Lightweight editor, not an IDE

Pi-Harness edits readable text with lazy syntax highlighting, line numbers, undo/redo, find, explicit save, unsaved-state indicators, and external-change conflict protection. Oversized, binary, media, and document files use read-only previews.

It deliberately does not include LSP/IntelliSense, semantic refactoring, a debugger, task runner, integrated terminal, or IDE extension compatibility.

Architecture

Pi-Harness separates the desktop control plane from the growing visual Harness Console while keeping Pi Coding Agent as the only Agent Runtime.

                         Pi-Harness

             ┌──────────────┴──────────────┐
             │                             │
       Control Plane                 Harness Console
             │                             │
        Providers                      Overview
        Models                         Runs
        Skills                         Trace
        Packages                       Context
        Environment                    Tools
        Config                         Policy
        Updates                        Checkpoints
        Backups                        Evaluation
        Diagnostics                    Sessions
                                       Stats
             │                             │
             └──────────────┬──────────────┘
                            ▼
                    Pi Agent Harness
                            │
                            ▼
                    Pi Coding Agent
                            │
                            ▼
                         Models

The Control Plane manages everything around Pi. The Harness Console observes and controls Pi Agent Harness state: it records Runs, enforces Policy at the Pi tool boundary, creates Checkpoints, and evaluates finished runs from real evidence. Pi remains the only Agent Runtime; sessions stay compatible with the Pi CLI JSONL under ~/.pi/agent/sessions/.

Current screenshots

The same six product surfaces are shown in both the default and Classical Chinese themes.

Default theme

Workspace	Git

Providers	Models

Capabilities	Preferences

Classical Chinese theme

Workspace	Git

Providers	Models

Capabilities	Preferences

Coming Next

Pi-Harness has shipped its Harness Control Plane: Runs, Policy, Checkpoints, Evaluation, and a filterable Trace are live in the Harness