hermes - 💡(How to fix) Fix Feature Proposal: Background multi-agent harness (Doer/Reviewer + Hindsight shared memory)

Official PRs (…)
ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

Helpful · Quick feedback

Loading…
RAW_BUFFERClick to expand / collapse

Problem

There is currently a gap between delegate_task (synchronous, blocks parent conversation) and kanban (persistent, durable, audit-trailed) for background multi-agent workflows that benefit from lightweight review automation but do not need a full persistent board.

Common scenario: spawn 2-3 agents to research/download/verify in parallel, automatically review their outputs, and get results pushed back asynchronously without blocking the main conversation.

What we built

Orchestrator v3 — a background multi-agent harness (~850 lines) that sits between delegate_task and kanban:

  • Doer / Reviewer dual-role: Doers execute tasks in parallel; Reviewers automatically verify outputs. Strong role separation instead of homogeneous swarm workers.
  • Async non-blocking: Agents run in background; main conversation continues uninterrupted.
  • Hindsight real-time shared memory: Agents share findings via Hindsight semantic search (recall/retain), not static MEMORY snapshots or JSON blackboards.
  • L1 flash → L2 pro automatic review escalation: Flash reviewer runs first (~40s, low cost). Only escalates to Pro reviewer on failure, controlling cost.
  • Dynamic append: Add new tasks to a running session without restarting.
  • Hard timeout: Pro reviewer runs as subprocess with true 120s kill timeout.

Comparison with existing capabilities

Capabilitydelegate_taskkanban swarmOrchestrator v3
LifecycleSingle-shot syncPersistentBackground async
Blocks parentYesNoNo
Shared memoryWeakBlackboard/commentsHindsight real-time
Role modelSub-agentworker/verifier/synthesizerDoer/Reviewer
Auto review escalationNoneGated verifierL1 flash + L2 pro
Crash recoveryNoneStrongWeak (tradeoff)
Dynamic task appendNoNoYes
Official integrationHighHighLow (external)

Why this may matter to Hermes

This is a complementary workflow shape, not a replacement. It fills the gap for users who need background parallel execution, automatic review without manual verifier config, shared agent memory across workers, and cost-controlled escalation.

Question to maintainers

If this belongs upstream, should it be positioned as:

  1. A delegate_task profile/background enhancement?
  2. A kanban lane/helper under the external worker lane contract?
  3. An official optional-skill/plugin example?
  4. Or best kept as community-maintained external tooling?

Current limitations (honest)

  • Requires memory.provider: hindsight
  • No crash recovery (intentionally lightweight)
  • External script, not yet a Hermes plugin
  • Solvable engineering issues if direction is approved (13 P0/P1/P2 reliability issues already fixed)

Related upstream discussions

  • #4949 (background subagents)
  • #9459 (delegate_task profile/harness)
  • #19931 (kanban worker lanes)
  • #20157 (persistent profile-agent orchestration)

Vote matrix · Quick signals

Works
Did the solution work? Tap to confirm.
Easy Fix
Was it a quick fix?
Time Saver
Did it save you time?
Blocking
Was it severely blocking?
Common Issue
Are others likely hitting this too?
Flaky / Intermittent
Is it intermittent?
Verified / Reproducible
Can you reproduce it reliably?
Loading…

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Back to top recommendations

TRENDING

hermes - 💡(How to fix) Fix Feature Proposal: Background multi-agent harness (Doer/Reviewer + Hindsight shared memory)