AGENT SKILL

multi-bot-desk-orchestrator

Orchestrate multi-bot agent teams with per-bot serial turn queues, bounded hop-depth guards, in-band MCP tool approvals, and isolated display/gateway leases. Do not use for single-bot stateless chat or raw VNC tunneling.

License 许可
MIT
最近更新
2026-11-15
Stars
★ 0
Forks
0

01 前言 Front Matter

name
multi-bot-desk-orchestrator
description
Orchestrate multi-bot agent teams with per-bot serial turn queues, bounded hop-depth guards, in-band MCP tool approvals, and isolated display/gateway leases. Do not use for single-bot stateless chat or raw VNC tunneling.

02 资源文件 3

03 SKILL.md 正文

Multi-Bot Desk Orchestrator

multi-bot-desk-orchestrator distills the multi-agent desktop orchestration architecture from bot-hq. It governs how multiple persistent bots collaborate on a shared host without colliding on browsers, corrupting primary agent configs, entering infinite bot-to-bot loops, or bypassing human tool approvals.

See CARD.md for the companion summary card and mandatory invariants checklist. Consult config.default.yaml for baseline operational budgets and ../agent-harness-architect/SKILL.md for harness isolation design.


1. Per-Bot Serial Queue & Hop-Depth Guard

Serialize turns per bot_id and bound multi-bot chatter:

  1. Serial Turn Queue: Execute one turn per bot at a time so browser/desktop state never races.
  2. Hop-Depth Cap: Track current_depth across send_to_bot relays and reject dispatches once current_depth >= max_depth (default 3).
  3. Room Fan-Out Bound: In group rooms, route to @mentioned members or cap unmentioned broadcasts at 3 members.

Completion criterion: BotHopDispatchOutcome returned with bounded next_depth and target_bot_ids.


2. In-Band MCP Approval Gate

Enforce human-in-the-loop safety inside the MCP tool endpoint:

  1. Read-Only Fast Path: Read-only tools and explicitly always_allowed_tools execute immediately (auto_allow).
  2. Blocking Write Card: Mutating tools suspend the turn (require_approval) until the user approves or refuses.
  3. User Interrupt Supersession: A new user message immediately supersedes pending approval cards as skipped.

Completion criterion: McpApprovalOutcome classifies every tool call without engine-specific forks.


3. Primary-vs-Managed Home & Screen Isolation

Isolate managed bots while preserving the host's primary agent:

  1. Primary Non-Interference: Never restart, stop, or rewrite the primary bot's config (mutates_primary_config = False).
  2. Dedicated Resource Leases: Assign each managed bot a unique display_index (1–3), gateway_port (8650 + slot), and cdp_port (9200 + slot).

Completion criterion: BotDeskLease allocated with zero port or display collisions.


The Three Foundational Pillars

1. Bounded Multi-Bot Collaboration

Cap bot-to-bot hops at 3 and room fan-outs at 3 so teams collaborate without runaway ping-pong loops.

2. In-Band MCP Safety Boundary

Gate mutating tools inside the MCP endpoint so every channel and engine shares one approval contract.

3. Non-Destructive Primary Coexistence

Run managed bots in isolated homes, ports, and displays without mutating the primary host agent.


Anti-Patterns

  • Unbounded Bot Ping-Pong — Allowing send_to_bot relays without a strict max_depth ceiling, causing infinite turn loops.
  • Shared Browser Helper Collision — Launching multiple managed bots against the same BH_HOME or Chrome DevTools port.
  • Primary Config Clobbering — Restarting or overwriting the host's primary agent config.yaml when adding companion bots.
  • Stale Approval Deadlock — Leaving a pending tool approval card active after the user has already sent a new message or interrupted the turn.

04 同仓库更多 Skills

SpectreDeath
SpectreDeath/Brain-Harness

ui-behavior-primitive-architect

Architect, audit, and verify zero-dependency headless UI behavior primitives, WebMCP agent bridges, WAI-ARIA/SSR/RTL tri-modal checks, and framework expert routing. Do not use for static CSS theme styling or backend database migrations.

★ 0
SpectreDeath
SpectreDeath/Brain-Harness

trustgraph-ontology-governor

Govern knowledge graphs, SHACL-AF triple policies, OntologyRAG subclass validation, PROV-O dual named graphs, and Laya entropy decision scoring. Do not use for relational SQL queries or generic web scraping.

★ 0
SpectreDeath
SpectreDeath/Brain-Harness

shim-harness-governor

Govern universal CLI agent loops, 4-tier tool permissions, ancestor-symlink workspace sandboxes, deterministic-first command routing, collision-safe MCP namespacing, and layered skill loading. Do not use for static UI styling or external SQL analytics.

★ 0
SpectreDeath
SpectreDeath/Brain-Harness

graphiti-zep-memory-governor

Govern bi-temporal knowledge graphs with Graphiti future-edge clamping, MinHash/LSH entity dedup, RRF/MMR reranking, and Zep typed ontologies. Do not use for static SQL joins or generic web crawling.

★ 0
SpectreDeath
SpectreDeath/Brain-Harness

agora-mcp-workbench-architect

Architect, audit, and orchestrate Agora Workbench MCP servers, loopback sidecars, unified state graphs, gateway/dispatcher connectors, and policy-enforced data lakes. Do not use for static frontend CSS styling or non-MCP desktop GUI automation.

★ 0
SpectreDeath
SpectreDeath/Brain-Harness

skill-eval-assurance-runner

Evaluate and assure agent skills using small-N Wilson confidence intervals, unbiased pass^k reliability, deterministic trajectory graders, preflight boundary honesty, and sealed digest verification. Do not use for subjective prose review or unconstrained production deployment.

★ 0