fix: 多 Agent 路由场景下按当前 agent 读取 footer metrics

[babobi-bo]

背景

修复 #502。

Fixes #502.

在 OpenClaw 多 Agent 路由场景下，飞书/Lark 消息已经能正确路由到非 main agent，例如 coder，但 streaming/card footer 里的模型/运行时指标可能不显示。

根因是 footer metrics 查询 session store 时，没有把当前路由到的 agentId 传给 resolveStorePath(...)，导致它默认去读 main/default agent 的 session store。对于被路由到非 main agent 的会话，模型和 token metrics 实际写在对应 agent 的 store 里，因此 footer lookup 读不到 metrics.model，最终 model footer 被跳过。

修改内容

本 PR 将当前路由的 agentId 传入 StreamingCardController，并在 footer metrics 的两条 session store 解析路径里使用它：

• runtime.agent.session.resolveStorePath(...)
• runtime.channel.session.resolveStorePath(...)

涉及文件：

• src/card/reply-dispatcher-types.ts
  • 给 StreamingCardDeps 增加 agentId
• src/card/reply-dispatcher.ts
  • 创建 StreamingCardController 时传入当前 agentId
• src/card/streaming-card-controller.ts
  • 调用 resolveStorePath(sessionStorePath, { agentId: this.deps.agentId })

验证

本地已验证：

npx pnpm typecheck
npx pnpm test
npx pnpm build

结果：

• TypeScript typecheck 通过
• 测试通过：27 个 test files / 214 个 tests
• build 通过

另外也已把相同修复临时覆盖到本地运行中的 @larksuite/openclaw-lark runtime 包进行验证，确认 footer/model 显示功能正常。

[evandance]

Thanks for the well-scoped bug report in #502 and this targeted fix — root cause analysis was spot on, and the implementation follows the same pattern already used in src/card/tool-use-config.ts:60, which keeps it consistent with how the codebase handles routed agentId in session-store lookups. Verified the SDK supports passing { agentId } to resolveStorePath (plugin-sdk/src/config/sessions/paths.d.ts:19), all call sites of createFeishuReplyDispatcher are updated (dispatch.ts:256, dispatch-commands.ts:67, the test setups), and pnpm test passes 214/214.

One non-blocking suggestion: there's no regression test pinned to the new opts-passing behavior. A test that mocks runtime.agent.session.resolveStorePath and asserts the opts argument carries { agentId: <routed-agent-id> } would prevent a future refactor from silently dropping it.

LGTM, approving.