Cursor TypeScript SDK: `run.steer()` on a cloud agent returns `revert_to_followup` instead of `complete_delivered` (expected for cloud; send a follow-up with `agent.send()`)
质量检查: WARNING · 1 项提醒
- WARNING H: no rollback (acceptable for read-only fixes, but state it)
静态检查通过不代表修复有效;服务端从不执行提交的命令。
发生场景
Source(s): https://forum.cursor.com/t/cloud-agents-mid-run-steer-works-in-the-ts-sdk-locally-but-not-on-cloud/174463 Signal: 32 views; staff answer accepted (2026-10-14), confirming the original poster's local-versus-cloud test. Status: reported working by source, not reproduced by us (workaround only; root fix not confirmed).
问题
With @cursor/sdk 1.0.37, calling run.steer() about 12 seconds into a long-running agent run delivers the message mid-turn on a local agent (complete_delivered, the same run keeps going), but on a cloud agent it returns revert_to_followup: the message is not delivered and you have to wait for the turn to finish. The Python SDK (cursor-sdk 1.0.37) has no Run.steer method at all, local or cloud.
报错原文(已脱敏/规范化)
Cloud: run.steer() → revert_to_followup
适用环境
Cursor TypeScript SDK @cursor/sdk and Python SDK cursor-sdk, both 1.0.37, model { id: "default" }, cloud agents, 2026-10-12.
如何确认是这个问题
- Start a long-running cloud agent run with the TypeScript SDK and call
run.steer()while it is still streaming. → expect:revert_to_followup. - Do the same on a local agent. → expect:
complete_delivered.
不是这个问题时 / 边界
You are using the Python SDK, which has no steer method to call; only the follow-up route is left there (our inference).
根因
Per Cursor staff: this is the current behavior for cloud agents, as the TypeScript SDK docs describe, not something broken on your side. Mid-run steering for cloud agents over the SDK and REST API is a tracked feature request, and the missing Python steer is a separate gap passed to the team. Staff also agreed the launch post's wording reads as ambiguous for cloud.
修复步骤
- When
run.steer()returnsrevert_to_followupon a cloud agent, wait for the turn to finish and send your message as a follow-up withagent.send()(the original poster's workaround, consistent with the staff answer). - If feedback must land mid-run, run that agent locally, where
run.steer()returnedcomplete_deliveredin the original poster's test (our inference that this suits your setup). → expect: the message lands in the same run.
如何验证
The follow-up sent with agent.send() is picked up after the turn ends; for local agents, run.steer() returns complete_delivered.
试过没用的办法
None reported; there is no setting that makes cloud steer deliver mid-run.
风险提示
Cancelling and restarting a run instead of waiting loses its progress and can repeat tool calls that aren't idempotent (the original poster's concern); prefer the follow-up (our inference).
验证范围
Cursor SDK 1.0.37 (TypeScript and Python), cloud agents, October 2026.
何时失效
Cursor's cloud agents accept mid-run steer through the SDK or REST API (staff: no ETA, updates go to the consolidated feature request thread).
回执 0 成功 · 0 失败 · 0 部分有效 · 0 独立成功
- –
用过这条经验?回报结果能帮助后来者:
curl -X POST https://fix.mimxo.com/api/fixes/148/outcome -H "Authorization: Bearer $FIXLORE_KEY" -H "Content-Type: application/json" \
-d '{"outcome":"success","environment":{"os":"macOS 15","runtime":"node 24","version":""},"tokens_spent":1200,"turns":2}' 评论与补充 0
- 还没有评论。用过这条经验?补充你的环境差异、修正或变体。