skills.rest
Kingxiao
KingxiaoCOMMUNITY
Explore all skills by this author →
GitHub

agentforge-harness

Apply harness-level constraints, hooks, and verification loops to recurring agent failures.

Updated Apr 8, 2026
Search Tags:#hooks#multi-agent#ci-cd#claude-code#harness-engineering#hashimoto-loop
One-click install
npx skills add https://github.com/Kingxiao/agentforge --skill agentforge-harness
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: agentforge-harness
Source: https://github.com/Kingxiao/agentforge/tree/main/agentforge-harness
Command: npx skills add https://github.com/Kingxiao/agentforge --skill agentforge-harness

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve?

Harness Engineering provides a disciplined, repeatable framework to prevent AI agent failures by encoding constraints, tests, and verification into CLAUDE.md rules, hooks, and a shared governance model.

Core Features & Use Cases

  • Hashimoto Loop implementation: observe failures, diagnose, classify fixes (behavioral, mechanical, structural, or context), and apply patches with verification.
  • Seven-layer harness architecture: Context Engineering, Tool Orchestration, Memory & State, Architectural Constraints, Verification & Feedback, Entropy Management, and Human-in-the-Loop.
  • Long-running and HTTP service harness patterns: heartbeat, checkpointing, auto-resume, health checks, idempotency, and graceful shutdown.
  • Team harness collaboration and multi-agent coordination: git-backed memory, CLAUDE.md governance, shared hooks, and PR-review discipline.
  • Self-evolution concepts: trajectory capture, safety gates, and patch-based skill improvements under user approval.

Quick Start

Start by cloning the harness repository and running the CLAUDE.md-driven setup to begin diagnosing and patching agent failures.

Frequently Asked Questions about agentforge-harness

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

FAQPage Schema
How do I prevent recurring AI agent failures during long-running operations?▼

Prevent recurring AI agent failures by applying harness-level constraints, mechanical hooks, and verification loops via CLAUDE.md rules. This framework implements heartbeat, checkpointing, and auto-resume patterns to ensure stability during long-running operations.

What is the Hashimoto Loop for agent failure diagnosis?▼

The Hashimoto Loop is an agent failure diagnosis process where you observe failures, diagnose root causes, classify fixes as behavioral, mechanical, structural, or context-based, and apply patches with verification to ensure stability.

How do I coordinate multi-agent systems using CLAUDE.md governance?▼

Coordinate multi-agent systems by establishing git-backed memory, shared hooks, and PR-review discipline within a CLAUDE.md governance model. This team harness collaboration framework ensures structured progress tracking and consistent multi-agent coordination.

Does harness engineering work with Claude Code for CI-CD pipelines?▼

Yes, harness engineering works with Claude Code by encoding constraints, tests, and verification into CLAUDE.md rules and hooks. This integration applies across CI-CD phases from prompt discipline to multi-agent coordination and team collaboration.

What are the limitations of using hooks for AI agent stability?▼

Hooks for AI agent stability require a CLAUDE.md-based rule set and structured progress tracking to validate fixes before deployment. Without mechanical hooks and verification loops, the harness cannot effectively classify and patch behavioral or structural agent failures.

Related Skills

Explore More Software Engineering Skills→
Fantasia1999Fantasia1999

review

Analyze AI memory logs for promotion, stale entries, and merging opportunities.

Community
Advanced
farmountainfarmountain
1

DevOps & Platform Domain Expert

Define deployment models, CI/CD gates, and rollback strategies for software projects.

Community
Intermediate
sakamoto-family-smilesakamoto-family-smile

dmux-workflows

Coordinate parallel AI agent sessions across multiple harnesses with dmux.

Community
Advanced
DanielMSchmidtDanielMSchmidt

ballroom-flow-validation-and-qa

Validate Weave Steps test routines for layer ownership and TDD discipline.

Community
Advanced