What problem does it solve? Verifying that every server-advertised official model actually completes through each supported external CLI agent (Claude Code, Codex, Grok Build, Kimi Code, Pi, TRAE) requires tedious manual testing across the real Desktop renderer, IPC, and relay path. This Skill automates that compatibility matrix with an executable harness and Acceptance-compatible evidence reporting. ## Core Features & Use Cases - Live Matrix Discovery: Reads the live getServerDefaultHeterogeneousCapability response and probes installed CLI binaries without making model calls, so the plan always matches the server's advertised agents and models. - Gated Live Execution: Runs each agent/model cell sequentially through the real server-default provider binding only after explicit --confirm-live approval, verifying a unique marker round trip plus authoritative relay, provider, and model attestation. - Acceptance-Compatible Reporting: Writes result.json, per-cell text evidence, and a structured matrix table under .records/reports/ for publishing as an immutable Acceptance round. - Use Case: Before a LobeHub release, invoke the skill to prove that every official model completes through Claude Code, Codex, and the other supported CLIs, then publish the resulting compatibility matrix as release evidence. ## Quick Start Ask the AI to run the testing-heterogeneous-agents compatibility matrix against the signed-in Electron instance and approve the discovered agent and model scope before any live requests are made.
