Skip to content

[bug] /idd-verify Devil's Advocate Agent role socket crash — 2x consecutive timeouts in #119 round 2 (closing follow-up from #119) #130

@kiki830621

Description

@kiki830621

Problem

/idd-verify 的 Devil's Advocate Agent role 在 #119 round 2 verify 期間連續 2 次 crash with API socket error:

  • 第 1 次 instance:API Error: The socket connection was closed unexpectedly after 462s (7.7 min)
  • 第 2 次 retry instance(per Recovery Protocol Step 2.5b):同 socket error after 420s (7 min)

兩次都在 polling loop + reading 4 sibling findings files + drafting DA review 期間 hit timeout / network drop。Other 4 reviewers (requirements / logic / security / regression) + Codex 都正常完成 — 只有 DA role hit 此問題。

Per idd-verify skill Step 2.5c,coordinator self-reviewed as fallback,但 quality lower than independent reviewer + 引入 verify engine degradation。

Type

bug / infrastructure(verify engine robustness)

Hypothesis(待驗證)

可能原因:

  1. DA role prompt 長度 — DA prompt 比其他 reviewer 多含 polling loop bash + 4 sibling findings cross-reference instructions,total context 可能比 other roles 大 2-3x
  2. Polling loop duration — DA 先 sleep + read 4 sibling files 後才開始 review,total agent-active time 比 other roles 長,增加 socket timeout window
  3. Multi-file Read pattern — DA 需 Read 4 sibling files + diff + issue body,Read-heavy 工作模式可能 trigger 某 API path 的 connection management 問題

Expected

DA Agent role 應該跟其他 reviewer 一樣 reliable;或 verify engine 應有更好的 fallback:

  • DA-specific timeout 設長(目前可能用 default 7-8 min)
  • DA prompt slimming(less context per call)
  • Or: split DA into 2 phases — quick poll + read,then separate adversarial review call

Actual

兩次 instance 都 socket crash,coordinator fallback 觸發 — verify engine degrade 為 5-AI + 1 self-review,而非 6-AI ensemble。

Source

surfaced during /idd-close #119 closing summary scan (Step 3.5) — round 2 verify process gap noted at comment-4507191109

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions