--- okf_version: "0.1" bundle: aria-multi-agent-orchestration source_bundle: "multi-agent-orchestration" title: "OKF Knowledge Bundle — Multi-agent orchestration & deliberation" updated: 2026-07-05 concepts: 6 --- # Multi-agent orchestration & deliberation — OKF v0.1 Knowledge Bundle Hierarchies, meta-harnesses, council/STORM deliberation and loop-of-loops control patterns for coordinating many agents. Self-contained, public-source AI-engineering material — no employer or personal data. Each entry is one concept (one markdown file); the `type` tag mirrors its role in the source encyclopedia. --- ## Concepts - [Claude Code Agent Hierarchies (agents.v1)](./claude-code-agent-hierarchies.md) · `concept` SteppieD's agents.v1: a 3-tier Claude Code hierarchy — 1 orchestrator, 7 team managers, 19+ specialists — routed by trigger keywords in agent descriptions. - [DeepResearchAgent (AgentOrchestra) — Hierarchical Multi-Agent Research System](./deepresearch-agent-orchestration.md) · `tool` SkyworkAI's hierarchical multi-agent framework (AgentOrchestra) tops GAIA (83.39%) with a planning conductor, execution agents, and a dynamic MCP Manager. - [LLM Council — Multi-Model Deliberation with Role-Assigned Members](./llm-council-multi-model-deliberation.md) · `technique` A council of role-assigned LLMs answers independently, blind-ranks each other, and a chairman model synthesizes — winning only on open-ended design questions. - [Loop of Loops — Recurring Agent Jobs as a Control Pattern](./loop-of-loops-control-pattern.md) · `talk` Nate B Jones' taxonomy: a prompt is one request, a loop is a recurring job with memory, and a loop of loops is a control pattern coordinating several loops. - [OmniGent — A Meta-Harness for All Your AI Agents](./omnigent-meta-harness.md) · `tool` Databricks' Apache-2.0 meta-harness sits above Claude Code, Codex, and Pi so they share one session, one history, and cross-vendor code review. - [STORM — Multi-Perspective Research, and a Claude-Skill Adaptation](./storm-multi-perspective-research.md) · `technique` Stanford's STORM writes cited Wikipedia-style articles via discovered multi-perspective research; a Claude skill borrows the thesis with five fixed personas. *Consumer: `python okf-cli.py list | search "query" [--semantic] | read `. See [README](./README.md).* --- --- theme: agents type: meta status: synthesized updated: 2026-07-05 source_bundle: "multi-agent-orchestration" --- # Multi-agent orchestration & deliberation — OKF v0.1 Knowledge Bundle ## What this is A self-contained, shareable **OKF v0.1 Knowledge Bundle** promoted from the ARIA-RD encyclopedia's **`multi-agent-orchestration`** logical bundle: Hierarchies, meta-harnesses, council/STORM deliberation and loop-of-loops control patterns for coordinating many agents. It gathers 6 public-source AI-engineering concept pages plus an index and a change log. Everything here is built from **public / external sources** — there is **no employer-specific or personal content**. ## Layout ``` index.md catalog / entry point (source_bundle declared in frontmatter) README.md this file log.md change log okf-cli.py dependency-free consumer (stdlib; BM25 + optional semantic search) *.md 6 concept pages (flat, bundle root) ``` Each concept is one markdown file whose id is its filename minus `.md`. Every non-reserved `.md` carries a `type` field in its frontmatter — the OKF conformance contract. `index.md` and `log.md` are reserved (excluded from search). ## How to use ```bash python okf-cli.py list # concept ids python okf-cli.py index # the annotated catalog python okf-cli.py search "your query" # BM25-ranked search with snippets python okf-cli.py search "your query" --semantic # embeddings + cosine (needs Ollama) python okf-cli.py read # print a concept python okf-cli.py selftest # OKF-conformance check ``` `search` ranks with BM25 over titles/descriptions/bodies. `--semantic` uses the precomputed `embeddings.json` (if present) and a local Ollama for the query vector, degrading gracefully to BM25 when either is unavailable. The legacy `find` keyword command is retained for backward compatibility. Cross-links between the 6 concepts are bundle-relative (`./slug.md`), so the bundle is navigable offline. ## External references These pages live in the full ARIA-RD encyclopedia (not in this bundle); inline links to them were unwrapped to plain text: - `Claude Code Agents v1 - Working Principles & Technical Architecture` - `agent-skills-progressive-disclosure` - `ai-native-company-operating-system` - `ai-native-organization-people-agents-context` - `anthropic-agent-sdk-billing-management` - `aria-subscription-not-api-economics` - `claude-code-observability-command-center` - `claude-fable5-youtube-content-pipeline` - `claude-haiku-45-capabilities` - `claudecodeui-remote-interface` - `context-engineering` - `deepresearchagent-MCP agent-guide` - `gas-town-agent-orchestrator` - `gstack-claude-code-engineering-team` - `guardrails-ai-validation` - `hermes-agent-goal-long-horizon-loop` - `hermes-agent-mcp-server-claude-code` - `linkedin-ai-agent-company-structure` - `local-ai-packaged-docker-stack` - `mcp-and-a2a-interop` - `oh-my-claudecode-orchestration` - `open-skills-portable-agent-procedures` - `pai-life-operating-system` - `personal-ai-computer-stack` - `rag-pipeline-architecture` - `ragas-evaluation` ## Provenance & interface Promoted from the **ARIA-RD** R&D AI-knowledge encyclopedia on **2026-07-05** from the `multi-agent-orchestration` logical bundle. `source_bundle: "multi-agent-orchestration"` (in this README and `index.md`) is the **interface contract** a nightly sync reads to keep this shipped bundle aligned with the logical bundle's live membership. This bundle is a copy — the encyclopedia remains the source of truth. See `log.md` for history.