Reconnect pull-wake stream after sleep

[KyleAMathews]

Detect long heartbeat timer gaps in the pull-wake runner and reconnect the wake stream after a computer resumes from sleep or a long event-loop stall. This keeps desktop agents listening for wake notifications instead of staying attached to a stale long-lived stream.

Root Cause

The pull-wake runner already reconnects its wake stream after repeated heartbeat failures. After a laptop sleep/resume, though, the long-lived wake stream can become stale while new heartbeat requests succeed again. In that state the runner can continue reporting a connected stream but stop receiving wake notifications.

This PR is stacked on #4632, which preserves deferred same-stream wake delivery and fixes stale in-flight heartbeat ownership across runner restarts. This change handles a separate failure mode: the process was suspended long enough that the existing wake stream should be considered suspect even if the next heartbeat succeeds.

Approach

Track wall-clock time between heartbeat timer ticks:

const elapsedMs = now - lastHeartbeatTickAt

If the elapsed gap is larger than resumeGapResetMs, the runner sends STREAM_RESET to the pull-wake lifecycle machine. The default threshold is the runner lease duration, so normal short event-loop delays do not reconnect the stream, while sleep/resume-sized gaps do.

The test simulates sleep by advancing wall-clock time before the heartbeat timer gets CPU again, then verifies that the current stream is cancelled and a new stream connection is opened.

Key Invariants

• Heartbeat requests still run on their normal cadence.
• Normal short timer drift does not reset the stream.
• Sleep/resume or long event-loop stalls force a wake-stream reconnect.
• The existing Refactor wake registry sync and fix child wake delivery #4632 guarantees around deferred same-stream wakes and heartbeat ownership remain intact.

Non-goals

• This does not change durable-stream client retry behavior.
• This does not change server-side runner leases or wake claim semantics.
• This does not replace heartbeat-failure-based reconnects; it adds a separate resume-gap signal.

Trade-offs

The runner may reconnect after any event-loop stall longer than the lease, not only literal OS sleep. That is intentional: after a lease-sized gap, the long-lived stream is cheap to re-establish and safer than assuming it is still delivering wake events.

Verification

cd packages/agents-runtime
pnpm exec vitest run test/pull-wake-runner.test.ts --reporter=dot
pnpm run typecheck

Also run from the repository root for changeset coverage:

GITHUB_BASE_REF=main node scripts/check-changeset.mjs

Verified locally:

• pull-wake-runner.test.ts: 65 passed
• @electric-ax/agents-runtime typecheck: passed after building the fresh worktree dependency @electric-ax/agents-mcp
• Changeset check: passed

Files changed

• packages/agents-runtime/src/pull-wake-runner.ts
  • Adds resumeGapResetMs configuration.
  • Tracks heartbeat timer wall-clock gaps.
  • Forces STREAM_RESET when the gap exceeds the configured threshold.
• packages/agents-runtime/test/pull-wake-runner.test.ts
  • Adds regression coverage for reconnecting after a simulated sleep/resume gap.
• .changeset/reconnect-wake-stream-after-sleep.md
  • Adds a patch changeset for @electric-ax/agents-runtime.

Stacking

Stacked on #4632.

[github-actions]

Electric Agents Desktop Builds

Build artifacts for commit ae585ea.

Platform	Status	Artifact
macOS Apple Silicon	Passed	DMG
macOS Intel	Passed	DMG
Windows x64	Passed	Installer
Linux x64	Passed	AppImage / deb

Workflow run

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 58.00%. Comparing base (2cc6c26) to head (ae585ea).
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@                     Coverage Diff                     @@
##           fix-child-wake-delivery    #4634      +/-   ##
===========================================================
- Coverage                    58.01%   58.00%   -0.02%     
===========================================================
  Files                          342      342              
  Lines                        40345    40364      +19     
  Branches                     11778    11779       +1     
===========================================================
+ Hits                         23407    23413       +6     
- Misses                       16900    16913      +13     
  Partials                        38       38              

Flag	Coverage Δ
packages/agents	72.64% <ø> (ø)
packages/agents-mobile	80.67% <ø> (ø)
packages/agents-runtime	83.49% <100.00%> (-0.05%)	⬇️
packages/agents-server	75.48% <ø> (-0.03%)	⬇️
packages/agents-server-ui	7.51% <ø> (ø)
packages/electric-ax	51.06% <ø> (ø)
typescript	58.00% <100.00%> (-0.02%)	⬇️
unit-tests	58.00% <100.00%> (-0.02%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[github-actions]

Electric Agents Mobile Build

Local mobile checks ran for commit ae585ea.

The EAS Android preview build was skipped because the mobile-eas-build label is not present.
Add the mobile-eas-build label to this PR to produce an installable preview build.

Workflow run