AGENT SKILL

kairo-test-harness

Orchestrates local gauntlet stress-testing, mock model setup, and memory benchmarking for the Kairo-Phantom digital copilot.

License 许可
MIT
最近更新
2026-07-18
Stars
★ 0
Forks
1

01 前言 Front Matter

name
kairo-test-harness
description
Orchestrates local gauntlet stress-testing, mock model setup, and memory benchmarking for the Kairo-Phantom digital copilot.

03 SKILL.md 正文

Kairo Phantom Test Harness

Overview

This skill encapsulates the workflow for testing, verifying, and certifying the Kairo-Phantom copilot daemon and agent squads. It automatically runs a 69-scenario parallel E2E gauntlet, performs the KMB-1 memory recall benchmark, manages background mock endpoints, and compiles final compliance reports.

Dependencies

  • None (self-contained CLI suite)

Quick Start

To set up, run the gauntlet, run the benchmark, and generate the final report:

uv run --directory C:\Users\praja\.gemini\config\plugins\science\skills\kairo-test-harness\scripts\ kairo_test_harness.py run-all --repo-path C:\Users\praja\OneDrive\Desktop\test-env\repositories\kairo-phantom --output C:\tests\results\master_run.json

Utility Scripts

The skill is backed by a multi-command helper script kairo_test_harness.py supporting the following subcommands:

1. setup

Starts the mock Ollama server and launches the Kairo-Phantom daemon process.

uv run kairo_test_harness.py setup --repo-path C:\Users\praja\OneDrive\Desktop\test-env\repositories\kairo-phantom

2. run-gauntlet

Executes all 69 scenarios across the 12 parallel squads using a configurable concurrency cap.

uv run kairo_test_harness.py run-gauntlet --repo-path C:\Users\praja\OneDrive\Desktop\test-env\repositories\kairo-phantom --max-parallel 4 --output C:\tests\results\gauntlet_results.json

3. run-benchmark

Executes the memory recall benchmark and checks if the composite score exceeds the threshold of 0.95.

uv run kairo_test_harness.py run-benchmark --repo-path C:\Users\praja\OneDrive\Desktop\test-env\repositories\kairo-phantom --output C:\tests\results\benchmark_results.json

4. cleanup

Termates any running background daemon and mock processes safely.

uv run kairo_test_harness.py cleanup --repo-path C:\Users\praja\OneDrive\Desktop\test-env\repositories\kairo-phantom

Common Mistakes

  • Port conflicts: Running the harness when another process is already listening on ports 7437 or 11435.
  • Wrong execution path: Running without setting up the workspace repository path. Specify the repository path using --repo-path if not executing from the workspace root.

04 同仓库更多 Skills