feat(remote_agent): add dispatch guard for queue load

[PiyushDatta]

Summary

• Apply WorkerPal completion ce445545-27c9-4d64-ae0c-5ebef047b619 to main_agents.
• Integrate commit 20909f6ba7ff04dec7ebcd604c7f3391a623946c from refs/pushpals/agent/workerpal-d4f80a1c/ce445545-27c9-4d64-ae0c-5ebef047b619.
• Worker workerpal-d4f80a1c reported: Executed task and modified 1 file(s) (quality gate soft-pass after 4 auto-revision attempt(s))
• Canonical task request: Implement an adaptive dispatch guard in apps/remotebuddy/src/autonomous_engine.ts that ingests queue telemetry (moving p95 latency, worker idle counts), throttles or reroutes delegation when pressure exceeds configured thresholds, and em...

Motivation / Context

• Preserve and review autonomous worker output before final merge to base branch.
• Keep integration branch current with queued worker completions.

Planned Scope

• apps/remotebuddy/src/autonomous_engine.ts

Planned Validation

• Planned: bun run test:root

Changes

• Updated apps/remotebuddy/src/autonomous_engine.ts

Testing / Validation

• Planned: bun run test:root
• Worker completion summary did not include explicit command pass/fail output.

Impact / Risk

• Risk level: medium (automated worker-generated change; maintainer review required).
• No secrets or credentials are expected in this PR body.

SourceControlManager Note

• Use this worker-provided PR title/body when creating the integration PR.
• Suggested title: fix(repo): Implement an adaptive dispatch guard in apps/remotebuddy/src/autonomous_engine.ts that ingest...

Checklist

• Tests added/updated where appropriate

• Validation commands run (or noted as not run)

• Docs/comments updated if needed

• No sensitive data (secrets/tokens) committed

• Agent branch: agent/workerpal-d4f80a1c/ce445545-27c9-4d64-ae0c-5ebef047b619

• Completion ref: refs/pushpals/agent/workerpal-d4f80a1c/ce445545-27c9-4d64-ae0c-5ebef047b619

• Commit: 20909f6ba7ff04dec7ebcd604c7f3391a623946c

• Completion ID: 073d0178-2e87-4c4c-8095-4300cc231b82

[PiyushDatta]

ReviewAgent: Changes Rejected (score 7.8/10)

Verdict: The dispatch guard implementation is well-structured and thoroughly unit-tested, but it contains a policy-breaking fallback that can bypass protection under queue pressure.

Issues:

• Policy bypass bug: when guard decision is reroute and no eligible queue_health candidate exists, code forces action: "allow" via queue_guard_reroute_override, which dispatches the original non-queue objective during elevated pressure and defeats enforceQueueHealthReroute semantics.
• Behavioral gap: no test covers the reroute-with-no-candidate path, so the unsafe fallback is unverified and currently unchecked by the suite.
• Operational risk: fetchQueueTelemetry() uses an unbounded fetch with no timeout/abort path, so a slow or hung /system/status call can stall objective dispatch flow.

This PR has been re-queued for automated fixes. A worker will address the issues above.

[PiyushDatta]

ReviewAgent: Changes Rejected (score 7.9/10)

Verdict: The dispatch guard feature is thoughtfully implemented with strong unit coverage, but it contains a policy bypass and operational resilience gaps that prevent production-ready confidence.

Issues:

• Policy bypass under elevated load: when guard action is reroute but no eligible queue_health candidate exists, the code creates queue_guard_reroute_override with action: "allow" and proceeds with normal dispatch; this violates strict queue protection intent and can still dispatch non-queue-health work during elevated pressure.
• No timeout/abort on fetchQueueTelemetry() (GET /system/status): a slow or hanging status endpoint can stall candidate processing in the autonomy tick and degrade scheduler responsiveness.
• Test coverage misses the most sensitive integration branch: there is no end-to-end test asserting behavior when reroute is required but unavailable (currently fail-open), and no test validating that blocked objectives are recorded with the expected status: "blocked" and block_reason for guard-throttle/reroute cases.

This PR has been re-queued for automated fixes. A worker will address the issues above.