KILab

KILab is a versioned coding-agent pack for AI-assisted software development, with optional local-model benchmarking material for GitHub Copilot, Claude Code, and Pi.

Quick start: coding-agent-pack

Clone the source pack and install all Copilot agents and skills into the current repository:

git clone https://github.com/franciscosuca/KILab.git "$HOME/kilab"

"$HOME/kilab/coding-agent-pack/scripts/install-pack.sh" \
  --target "$PWD" \
  --harness copilot \
  --agents all \
  --skills all

Install only selected resources with comma-separated names:

"$HOME/kilab/coding-agent-pack/scripts/install-pack.sh" \
  --target "$PWD" \
  --harness copilot \
  --agents test-oracle,blind-implementer \
  --skills feature-planning

See coding-agent-pack/README.md for Claude and Pi installation, project/global scope, updates, and validation.

Local model evaluation

The benchmark material is optional and independent from the coding-agent pack. It uses uv, LM Studio, and Docker for repeatable local model evaluations and sandboxed code execution.

What's included

• Inspect AI + LM Studio guide — benchmark setup and execution instructions.
• Benchmark results — recorded HumanEval results and throughput notes.
• Custom benchmark — a small repeatable VS Code/Copilot model comparison.

License

KILab is released under the MIT License.