Agent Lightning v1.0

Who Should Read

Backend developers who spend lots of time hand-tuning accuracy, cost, and latency of their own AI agents

Core Mechanics

• First official Skill release
  • v1.0.1 is the first official release of the Agent Lightning Skill.
  • The Skill helps coding agents optimize other AI agents.
• Input is an agent plus a benchmark
  • You provide an editable agent and a benchmark.
  • The Skill guides systematic improvements.
• Wide range of tuning targets
  • It covers prompts, tools, workflows, models, and reasoning settings.
  • It balances accuracy, cost, latency, and reliability.
• Measured iteration
  • Improvements proceed through measured iteration.
• Install for Claude Code, Codex, or Copilot
  • Install with the gh skill install command.
  • Pass the target agent with the --agent option.

Community Reactions

• Opinion that the README is odd
  • Commenters found the README's 3,500-line lightweight agentic RL framework pitch and 'simplicity as first principle' claim strange.
  • They felt the presentation was weak for a large company's project.
• Confusion about what the project is
  • One commenter understood it as a toolkit for further training or fine-tuning an LLM on agentic tasks.
  • That differs in flavor from this release's Skill.
• Question about effectiveness
  • A commenter asked how good it is.
  • The source gives no performance numbers.

How to Apply

• Install the Skill
  • Run the gh skill install command for the coding agent you use today.
  • Open your internal agent repo and invoke the Skill.
• Build the benchmark first
  • List 20-50 representative questions with correct-answer criteria.
  • Hand this set to the Skill as the improvement baseline.
• Rank your target metrics
  • State whether accuracy, cost, latency, or reliability comes first.
  • Record metric changes after each edit to decide whether to keep it.

Try It

# Run in terminal (install command from the source; replace <agent> with your agent, e.g. claude-code)

gh skill install microsoft/agent-lightning agent-lightning --agent <agent>

# Instruction to paste into your coding agent after install
# Replace {{agent path}} and {{benchmark path}} below
Use the Agent Lightning Skill to improve the agent in {{agent path}} against {{benchmark path}}. Vary prompts, tools, workflows, models, and reasoning settings, measure accuracy, cost, latency, and reliability, and summarize each change in a table.

# Next: compare scores before and after.

Terminology

Related Papers

• Same Simulation, Different Evidence: Where Retrieved Tool Output Costs LLM Agents Accuracy

  Saving long tool output to files cut GPT-5.1 accuracy by 11.3 points; naming the authoritative file restored it.

• Fake News, Real Crashes: Information Injection Attacks on LLM Financial Trading Agents

  Fake headlines hijacked LLM trading agents 62% of the time versus 14% for prompt injection.

• Harness Engineering 101: The LLM is the brain, the harness is the body

  An agent is a loop managing a JSON array; 16 chapters build one from scratch

• Harness2 by Google Research – Recursive Agent Harnessing

  Recursively refining the harness with fixed model weights raised benchmark scores by up to 11.8 points.

• Ousted Savannah police officer gave ICE agent husband access to Flock system

  A police officer shared her Flock login with her ICE-agent husband; an audit found 983 searches and she was fired

• Tmux and Claude Code: Sessions, Send-Keys, and the Ceiling

  Running Claude Code agents in tmux works well, but limits appear around four agents.

Related Resources

• https://github.com/microsoft/agent-lightning/releases/tag/v1.0.1