Darwin-AgentOfficial·12 Agent Skills Included

HarnessX

Composable harnesses for running, evaluating, and evolving agents

Builds and runs configurable agent harnesses with pluggable tools, memory, processors, and model providers. Captures reward-annotated trajectories from every run to feed SFT and RL training pipelines. Eliminates rewriting agents when swapping behaviors, models, or benchmarks. Includes benchmark playbooks, document skills, and a lab UI to test and evolve configurations faster.

415|49

↗

npx skills add Darwin-Agent/HarnessX --all -g -y

Available:

Points your AI coding assistant to the implementation guide that defines HarnessX's core architecture, canonical imports, composition patterns, and rules for building tools, processors, and memory backends.

All Skills in This Repository (12)

Pure Emerald Level Indicators

📦 In Repo

Darwin-Agent

gaia-playbook

Diagnose GAIA trajectory gaps and select intervention patterns for score improvements.

tb2-playbook

Define TB2 evolution constraints for sandbox topology, tool constraints, and trajectory signaling.

tau2-playbook

Automate τ²-bench evaluation setup and domain guidance for HarnessX.

journal

Store cross-round hypotheses and outcomes in YAML frontmatter for meta-agent workflows.

validate

Validate HarnessX artifacts with canonicalization, dry-fire, contract, and synthetic replay checks.

analyze

Analyze task trajectory frontmatter and body excerpts to generate candidate HarnessConfig lever hypotheses.

reference

Document HarnessX component authoring shapes and workflows for tools, processors, and prompts.

xlsx

Process spreadsheet inputs into formatted Excel workbooks with validation.

pdf

Extract text and data from PDF documents using Python libraries.

skill-creator

Automate Claude Skill creation, testing, and optimization with eval results.

pptx

Generate, edit, and template PPTX presentations from data sources.

docx

Automate Word document creation, editing, and formatting for .docx outputs.

Official

Advanced

Frequently Asked Questions

FAQPage Schema

How to install HarnessX?▼

Run `npx skills add Darwin-Agent/HarnessX --all -g -y` in your terminal to install all skills in this suite globally.

What does HarnessX do?▼

It is a Python framework for composing agent harnesses from reusable processors and tools, pairing them with any model, and collecting training trajectories from every run.

How does HarnessX improve agent performance?▼

A meta-harness analyzes run trajectories and automatically searches for better processor, prompt, and tool configurations, while reward-annotated data feeds RL fine-tuning of the model itself.

Which models and benchmarks does HarnessX support?▼

It works with providers like Anthropic and OpenAI via LiteLLM, and ships adapters for benchmarks such as GAIA, tau2-bench, and Terminal-Bench 2.

Can I use HarnessX skills without coding?▼

The document skills for PDF, Word, Excel, and PowerPoint run through plain-English requests, but the core harness framework is aimed at developers comfortable with Python.