Skip to content
Kitbay
Skill

testing-heterogeneous-agents

NOASSERTIONUpdated on Nov 11, 202684,974GitHub starsIncludes scripts

A maintainer skill in the LobeHub repository, run only when someone calls it, that checks whether six external command-line agents, among them Claude Code, Codex and TRAE, work through LobeHub's built-in official model provider in the desktop app.

What the skill teaches

A maintainer skill in the LobeHub repository, run only when someone calls it, that checks whether six external command-line agents, among them Claude Code, Codex and TRAE, work through LobeHub's built-in official model provider in the desktop app. A bundled Node harness first lists the matrix from the live capability response and the CLIs already installed, without calling any model; real runs make billable requests, so they need approval under the acceptance skill and an explicit confirm flag. Cells run one after another, each with a unique marker checked on the round trip, and results are saved as evidence files and published to LobeHub's acceptance records. Failures are diagnosed layer by layer.

Capabilities

  • List the agent and model test matrix without model calls
  • Run live round-trip checks per agent and model
  • Save evidence files for each test cell
  • Locate the failing layer of a broken cell