Head: fddeb88fefc789de365cac1a7d138664aeebeb9f (includes main44df336016)
Environment: fs-11 Linux x64; same real web dev server (7810) and server (15850); Claude Opus5.5; same throwaway git project.
Steps: fresh root-checkout Claude thread f0305734-2fb5-41fc-971f-df229ce13035; identical handoff prompt except new branch repro-25146-c (startFromOrigin=false, runSetupScript=false, continuationPrompt run pwd/branch then reply CLAUDE HANDOFF FINISHED).
Observed: workspace changed; cleanup interrupt SUCCESS, no new ProviderAdapterInterruptError.
Unchanged pre-existing behavior: the provider stream still fails during planned detach and continuation stays queued (separate issue22971); this change fixes only the subsequent closed-transport interrupt failure.
After interrupt span:
{
  "type": "effect-span",
  "name": "ClaudeAdapterV2.interruptTurn",
  "traceId": "abd8b0d4deb40ba69584fe45b2005e52",
  "spanId": "1144ece9f981fb51",
  "parentSpanId": "97e36ed7b711a554",
  "sampled": true,
  "kind": "internal",
  "startTimeUnixNano": "1792973444318479870",
  "endTimeUnixNano": "1792973444319104022",
  "durationMs": 0.624152,
  "attributes": {},
  "events": [],
  "links": [],
  "exit": {
    "_tag": "Success"
  }
}