harness-forge

Q: How to install harness-forge?

Create the skills directory: mkdir -p .claude/skills. Then add the config to .claude/skills/harness-forge.md. Finally, /harness-forge in Claude Code.

Q: What is harness-forge best for?

harness-forge is categorized under General.

New

56GitHub TrendingGeneralby 001TMF

Turn Claude Code into its own Meta-Harness — a skill that evolves the scaffolding around a fixed model (memory, retrieval, context, prompts) via a native propose→score→Pareto loop. Native reimplementation of Meta-Harness (Lee et al. 2026).

First seen 6/14/2026

Summary

This skill transforms Claude Code into a meta-harness that automatically evolves its own scaffolding—memory, retrieval, context, and prompts—through a propose-score-Pareto loop.

It reimplements the Meta-Harness approach (Lee et al.
2026), enabling developers to continuously optimize their AI workflow without manual tuning.

Install & Usage

Create the skills directory

mkdir -p .claude/skills

Download the skill file

Add the configuration to .claude/skills/harness-forge.md

Invoke in Claude Code

/harness-forge

Use Cases

Automatically refine the system prompt and context window strategy to improve response relevance for a specific project.

Evolve retrieval-augmented generation (RAG) parameters like chunk size and embedding model based on task performance feedback.

Optimize memory management policies (e.g., summarization triggers, forgetting curves) to balance recall and token usage.

Discover Pareto-optimal configurations for multi-objective trade-offs between response accuracy, latency, and cost.

Continuously adapt the skill's own scaffolding as the codebase grows, preventing degradation in assistant performance.

Benchmark and compare different prompt engineering strategies (e.g., chain-of-thought vs. direct answer) under real usage.

Usage Examples

/harness-forge propose --objective accuracy --constraints token_budget:4000

Run a Pareto optimization loop to find the best retrieval top-k and prompt template for my current repo.

Score the current harness configuration against a set of test queries and propose an improved variant.

View source on GitHub

Security Audits

LicenseUnknownSourceWarnRepositoryPass

Frequently Asked Questions

What is harness-forge?

This skill transforms Claude Code into a meta-harness that automatically evolves its own scaffolding—memory, retrieval, context, and prompts—through a propose-score-Pareto loop. It reimplements the Meta-Harness approach (Lee et al. 2026), enabling developers to continuously optimize their AI workflow without manual tuning.

How to install harness-forge?

To install harness-forge: create the skills directory (mkdir -p .claude/skills), then add the config to .claude/skills/harness-forge.md. Finally, /harness-forge in Claude Code.

What is harness-forge best for?

harness-forge is a community categorized under General. Created by 001TMF.

What can I use harness-forge for?

harness-forge is useful for: Automatically refine the system prompt and context window strategy to improve response relevance for a specific project.; Evolve retrieval-augmented generation (RAG) parameters like chunk size and embedding model based on task performance feedback.; Optimize memory management policies (e.g., summarization triggers, forgetting curves) to balance recall and token usage.; Discover Pareto-optimal configurations for multi-objective trade-offs between response accuracy, latency, and cost.; Continuously adapt the skill's own scaffolding as the codebase grows, preventing degradation in assistant performance.; Benchmark and compare different prompt engineering strategies (e.g., chain-of-thought vs. direct answer) under real usage..