raphaelchristiCommunity·9 Agent Skills Included

harness-evolver

Autonomous optimization of LLM agent code via LangSmith

Automatically improves LLM agent codebases by evolving prompts, routing, tools, and architecture through iterative experiments. Eliminates manual prompt tuning and guesswork by scoring every candidate against LangSmith datasets with LLM-as-judge evaluation. Runs proposers in isolated git worktrees, merges only verified winners, and blocks regressions with constraint and efficiency gates. Tracks progress with evolution charts, health checks, and certification runs for reliable, measurable quality

43|5

↗

npx skills add raphaelchristi/harness-evolver --all -g -y

Available:

Explains to the AI agent how the skills, sub-agents, and Python tools connect in the evolution loop, and how to invoke each LangSmith-backed tool when optimizing a user's agent codebase.

All Skills in This Repository (9)

Pure Emerald Level Indicators

📦 In Repo

raphaelchristi

dev:release

Automate npm package releases with changelog generation and GitHub publishing.

dev:dry-run

Smoke-test the evolve pipeline for tool syntax, argparse behavior, and end-to-end execution.

dev:validate

Validate Claude plugin versions, frontmatter, cross-references, Python syntax, and hooks.

harness:certify

Run repeated evaluations and compute score stability statistics for evolved LLM agents.

harness:deploy

Clean up artifacts, tag releases, and push to version control.

harness:health

Detect and auto-correct quality issues in LangSmith evaluation datasets.

harness:setup

Configure Harness Evolver, LangSmith, and baseline evaluations for Python projects.

harness:status

Display evolution progress and scores from the Harness Evolver workflow.

harness:evolve

Orchestrate propose-evaluate-iterate loops with LangSmith and Git worktrees.

Community

Advanced

Frequently Asked Questions

FAQPage Schema

How to install Harness Evolver?▼

Run `npx skills add raphaelchristi/harness-evolver --all -g -y` in your terminal to install all skills globally.

What does Harness Evolver do?▼

It autonomously improves your LLM agent's prompts, tools, and architecture by running iterative experiments scored through LangSmith, then merging only the changes that measurably win.

Do I need a LangSmith account to use it?▼

Yes. You need a LangSmith API key since all datasets, experiments, and LLM-as-judge evaluations run through LangSmith as the backend.

Which frameworks and editors are supported?▼

It works with any Python agent framework LangSmith can trace, including LangGraph, CrewAI, and OpenAI SDK, and runs in Claude Code, Cursor, Codex, and Windsurf.

Will it break my existing agent code?▼

No. Every candidate change is tested in an isolated git worktree, and only candidates that pass scoring, constraint, and efficiency gates get merged into your main branch.