heterogeneous-agent

Diagnose LobeHub heterogeneous agent integration failures across CLI trace, adapter mapping, and persistence stages.

1|Updated May 9, 2026
One-click install
npx skills add https://github.com/duwenji/generative-ai-oss-tutorials --skill heterogeneous-agent-duwenji
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: heterogeneous-agent
Source: https://github.com/duwenji/generative-ai-oss-tutorials/tree/main/sandbox/lobe-chat/.agents/skills/heterogeneous-agent
Command: npx skills add https://github.com/duwenji/generative-ai-oss-tutorials --skill heterogeneous-agent-duwenji

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Fixes confusing failures in LobeHub heterogeneous agent integrations (Claude Code, Codex, and future external CLI agents) when tool calls, step boundaries, or adapter event mapping go wrong.

Core Features & Use Cases

  • Root-cause the pipeline stage: validates CLI raw stdout/JSONL first, then adapter event mapping, then persistence and UI hydration.
  • Diagnose tool-call and boundary bugs: detects mixed multi-tool messages, broken turn/step boundaries, orphan tool messages, and stuck tool loading.
  • Enforce technical invariants: verifies tool-call id mapping, correct tool_result routing, subagent thread scoping, and resume/session handling behavior.
  • Use targeted materials: points you to the key controller, driver, adapter, and executor files plus a dedicated debug workflow reference.

Quick Start

Use this skill when your heterogeneous agent shows issues in the external CLI pipeline (e.g., mixed tools in one assistant, broken step boundaries, or tool_result not landing), then validate raw trace → adapted events → persistence invariants in that order.

Frequently Asked Questions about heterogeneous-agent

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I debug broken tool_result routing in LobeHub heterogeneous agent pipelines?

To debug broken tool_result routing in heterogeneous agent pipelines, validate CLI raw stdout/JSONL first, then trace adapter event mapping, and finally verify persistence invariants to isolate the failing stage.

Why does my Claude Code adapter show mixed multi-tool messages and broken step boundaries?

Mixed multi-tool messages and broken step boundaries in Claude Code adapters occur when adapter event mapping fails to correctly parse CLI raw JSONL events and enforce stable tool_call payload id invariants.

What is the correct order to troubleshoot Codex external CLI agent integration failures?

The correct order to troubleshoot Codex external CLI agent integration failures is to validate the CLI raw trace first, then check adapter event boundaries, and finally verify step-scoped message persistence and UI hydration.

How do I fix orphan tool messages and stuck tool loading in heterogeneous agent integrations?

Fix orphan tool messages and stuck tool loading by diagnosing adapter event mapping to ensure correct tool_call id mapping, subagent thread scoping, and proper tool_result resolution across the pipeline stages.

Does this debugging workflow apply to external CLI agents other than Claude Code and Codex?

Yes, this debugging workflow applies to any external CLI agent pipelines involving tool-call chaining, subagent thread routing, and resume/session handling regressions, including future integrations beyond Claude Code and Codex.

What are the limitations when diagnosing resume and session handling regressions in heterogeneous agents?

A limitation when diagnosing resume and session handling regressions is that the workflow targets adapter event boundaries and persistence invariants, meaning failures originating purely within the external CLI's internal state are not directly patched.