v0.24.11 -> v0.24.12
Summary
Dispatcher routing gains retry and the resolution step skips its probe when only one candidate remains, closing the single-agent failure mode where a transient backend hiccup left both the dispatcher call and the probe dead with no fallback left to try. Telegram and Discord runtimes surface the resolution error back to chat instead of leaving the user staring at silence while the daemon log carries the only trace.
翻譯
Dispatcher routing 補上重試、單候選 resolve 跳過 probe,修掉 v0.24.11 fallback 機制在單 agent setup 下被一次性 backend 抽風判死的失敗模式。Telegram 與 Discord runtime 在 resolve 失敗時改把錯誤回到 chat,user 不再只能盯著毫無回應的對話窗等 daemon log。
Changes
FEAT
- Add agent resolution retry and runtime error replies [1fa20c6]
翻譯
-
SelectAgentNamesdispatcher routing 加 3 × 10s retry loop,全失敗 return error 而非靜默 fall back(routing 失敗 ≠ agent 本身死,但連 routing 都 timeout 代表 backend 廣域不可用,繼續硬塞 registry order 只是浪費 user 下一輪 send timeout);ResolveAgent對len(candidates)==1直接 return primary,跳過 5s probe(單候選沒排序意義,probe 失敗等於把同一個 transient hiccup 雙重 gate);Telegram run.goResolveAgent失敗時FinishStatus+<blockquote>⚠️ ...</blockquote>HTML reply,Discord run.go 送⚠️ ...純文字,對齊execute.goMaxRetry exhaustedsendText + EventDone契約
Scope
-
internal/agents/exec/selectAgent.go— FEAT (dispatcher retry, skip-probe single candidate, propagate error) -
internal/runtime/telegram/run.go— FEAT (HTML error reply on ResolveAgent failure) -
internal/runtime/discord/run.go— FEAT (text error reply on ResolveAgent failure)
Generated by SKILL
Top comments (0)