Pause DAG runner idle timeout during SDK reconnect

[rajarshidattapy]

Summary

• The DAG runner treated SDK reconnect silence as a stalled model and marked the task ERROR.
• Stream idle waits now freeze while onConnectionStateChange reports reconnecting, and still honor the hard task deadline.
• The same change is mirrored in the copyable .cursor/skills/dag-task-runner skill.

Fixes #59

Test plan

• cd sdk/dag-task-runner && pnpm test (3/3)
• pnpm typecheck

---

Note

Medium Risk
Runner timing and global console patching affect all in-flight DAG tasks; mis-parsed SDK logs could leave idle timeout paused or tasks still failing on long retries, but behavior is covered by new tests and hard deadlines remain.

Overview
Fixes false ERRORs when the Cursor SDK retries transport stalls and the model stream goes quiet during reconnect.

Idle timeout behavior replaces the old fixed withTimeout race with withIdleTimeout, which only counts stream silence toward --stream-idle-timeout-ms when the runner is not in an SDK retry. The per-task deadline (--task-timeout-ms) still applies during retries. Stream iteration and post-stream wait() both use this helper.

Reconnect detection adds createSdkReconnectProbe, which temporarily wraps console.log/warn/info to infer retry state from SDK diagnostics (AGENT_ERROR_DIAGNOSTICS decision=RETRY/THROW, [nal_agent_retries] lines), keyed by originalRequestId for parallel tasks. Local agents are created with enableAgentRetries: true. @cursor/sdk is bumped to ^1.0.30 (lockfile slimmed; Node >=22.13 on the SDK).

Changes are mirrored in .cursor/skills/dag-task-runner; skill docs note that idle timeout pauses while reconnecting. Removes disable-model-invocation from the skill frontmatter. Adds pnpm test covering idle timeout and reconnect probe edge cases.

^{Reviewed by Cursor Bugbot for commit de8aa9e. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{Bugbot Autofix is ON, but it could not run because the branch was deleted or merged before autofix could start.}

^{Reviewed by Cursor Bugbot for commit e1d79ae. Configure here.}