Bug type

Behavior bug (incorrect output/state without crash)

Beta release blocker

No

Summary

The sessions_spawn tool in pi-embedded-BaSvmUpW.js hardcodes channel-specific logic (e.g., opts?.agentChannel === "wechat"), requiring code modifications for each new channel (feishu, telegram, discord, etc.). This violates the open/closed principle and duplicates channel plugin metadata that already declares threading capabilities.

Steps to reproduce

1. Configure Feishu channel in openclaw.json
2. Send message to Feishu bot: "帮我写个代码"
3. Main agent calls: sessions_spawn({ agentId: "coder", mode: "run" })
4. Without explicit thread: true, subagent creation fails
5. Current workaround requires hardcoding "feishu" check in core code

Expected behavior

The framework should automatically inject thread: true and mode: "session" for channels that declare this requirement in their plugin configuration, without hardcoding channel names in core code.

Main agent should only specify business logic (which expert to call), not technical implementation details.

Actual behavior

Current implementation (line 112131 in pi-embedded-BaSvmUpW.js):

const isFeishuChannel = opts?.agentChannel === "feishu";
const thread = params.thread === true || (isFeishuChannel && runtime === "subagent");

This requires:

Modifying core code for each new channel
Maintaining hardcoded channel name lists
LLM prompts must include technical details like "飞书渠道必须添加 thread: true"


### OpenClaw version

2026.3.24

### Operating system

Linux (tested on Ubuntu 24.04)

### Install method

npm install -g openclaw

### Model

local/qwen3.5:9b (main agent) coding-plan/glm-5 (Coder expert)

### Provider / routing chain

Feishu channel → Main agent (qwen3.5:9b) → sessions_spawn → Coder subagent (GLM-5)

### Additional provider/model setup details

_No response_

### Logs, screenshots, and evidence

```shell
Feishu enterprise self-built app
WebSocket event subscription enabled
dmPolicy: pairing
Subagents configured: analyst, coder, researcher, generalist, planner

### Logs, screenshots, and evidence

// Current hardcoded implementation (pi-embedded-BaSvmUpW.js:112131)
const isFeishuChannel = opts?.agentChannel === "feishu";
const effectiveMode = isFeishuChannel && runtime === "subagent" ? "session" : mode;
const thread = params.thread === true || (isFeishuChannel && runtime === "subagent");

// Error without thread=true:
{
  "status": "error",
  "error": "mode=\"session\" requires thread=true so the subagent can stay bound to a thread."
}

// Success with manual workaround:
{
  "status": "accepted",
  "childSessionKey": "agent:coder:subagent:xxx",
  "modelApplied": true
}

Impact and severity

Severity: Medium

Impact:

• Architectural debt - violates separation of concerns
• Maintenance burden - each new channel requires core code changes
• LLM confusion - models must remember channel-specific technical details
• Prompt pollution - SOUL.md cluttered with implementation details
• Not scalable - wechat, telegram, discord all need similar logic

Workaround exists but suboptimal (hardcoding channel names).

Additional information

Related files:

• packages/pi-embedded/src/pi-embedded-BaSvmUpW.js (line ~112131)
• packages/extensions/feishu/src/index.js (plugin definition)
• packages/plugin-sdk/src/channel-plugin.ts (plugin schema)

Channel plugins already declare capabilities.threads: true but this metadata is not used by sessions_spawn tool.

Similar pattern exists for replyToMode resolution which correctly reads channel plugin config.