{"id":51951728,"url":"https://github.com/dubsopenhub/hive1k","last_synced_at":"2026-07-29T06:01:29.282Z","repository":{"id":350617634,"uuid":"1207596449","full_name":"DUBSOpenHub/hive1k","owner":"DUBSOpenHub","description":"🐝 Hive1K — one thousand agents, one hive mind. Recursive multi-model swarm orchestration for the Copilot CLI. Launch 250–1,000+ AI agents across 16 models.","archived":false,"fork":false,"pushed_at":"2026-04-11T07:28:34.000Z","size":214,"stargazers_count":1,"open_issues_count":0,"forks_count":1,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-07-07T22:21:29.491Z","etag":null,"topics":["ai-agents","consensus","copilot-cli","copilot-extensions","github-copilot","multi-agent","multi-model","orchestration","prompt-engineering","recursive-swarm","shadow-scoring","swarm"],"latest_commit_sha":null,"homepage":null,"language":"Shell","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/DUBSOpenHub.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":"CONTRIBUTING.md","funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":"SECURITY.md","support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":"AGENTS.md","dco":null,"cla":null}},"created_at":"2026-04-11T06:15:58.000Z","updated_at":"2026-05-21T19:40:06.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/DUBSOpenHub/hive1k","commit_stats":null,"previous_names":["dubsopenhub/hive1k"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/DUBSOpenHub/hive1k","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/DUBSOpenHub%2Fhive1k","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/DUBSOpenHub%2Fhive1k/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/DUBSOpenHub%2Fhive1k/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/DUBSOpenHub%2Fhive1k/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/DUBSOpenHub","download_url":"https://codeload.github.com/DUBSOpenHub/hive1k/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/DUBSOpenHub%2Fhive1k/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":36020486,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-07-20T02:08:10.276Z","status":"online","status_checked_at":"2026-07-29T02:00:04.910Z","response_time":95,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["ai-agents","consensus","copilot-cli","copilot-extensions","github-copilot","multi-agent","multi-model","orchestration","prompt-engineering","recursive-swarm","shadow-scoring","swarm"],"created_at":"2026-07-29T06:01:27.488Z","updated_at":"2026-07-29T06:01:29.275Z","avatar_url":"https://github.com/DUBSOpenHub.png","language":"Shell","funding_links":[],"categories":[],"sub_categories":[],"readme":"# 🐝 Hive1K\n\n\u003e *One thousand agents. One hive mind.*\n\nYou type one sentence. Behind the glass, a thousand workers fan out across 16 models, argue in cross-family review meshes, get audited by sealed judges they never knew existed, and converge back into a single answer — usually before your coffee gets cold.\n\nIt’s quieter than you’d expect.\n\n[![License: MIT](https://img.shields.io/badge/License-MIT-green.svg)](LICENSE) [![Security Policy](https://img.shields.io/badge/Security-Policy-brightgreen?logo=github)](SECURITY.md)\n\n---\n\n## What happens when you type `hive1k`\n\n```text\n$ copilot\n\n\u003e hive1k h-500 \"Document the auth system, add missing tests, and flag rollout risks\"\n\n[NEXUS] Booting H-500 swarm...\n[NEXUS] Sealing acceptance criteria (8 checks)\n[NEXUS] Deploying 2 Division Commanders...\n[DIV-ALPHA] Commanding Architecture + Implementation divisions\n[DIV-BETA] Commanding Testing + Documentation + Integration divisions\n  [DIV-ALPHA/CMD-ARCH] Mapping auth boundaries and module ownership\n  [DIV-ALPHA/CMD-IMPL] Tracing token issuance, refresh, and revocation flows\n  [DIV-BETA/CMD-TEST] Enumerating missing happy-path, edge-case, and failure-path tests\n  [DIV-BETA/CMD-DOCS] Drafting operator-facing docs and examples\n  [DIV-BETA/CMD-INTG] Checking rollout risks across API, web, DB, and monitoring\n[REVIEW] Cross-family review mesh started\n[SHADOW] 1 criterion failed on first pass → hardening cycle triggered\n[SHADOW] Re-validated bundle: 0 critical failures remaining\n\n✅ Final bundle ready in 47s\n\nTop outputs:\n  1. Auth architecture brief with module boundaries\n  2. Ranked test-gap list with highest-risk paths first\n  3. Rollout checklist covering cookies, refresh tokens, and observability\n  4. Updated docs outline for onboarding + operations\n\nConsensus: CONSENSUS on 3/4 major findings\nShadow Score: 12.5% → hardened and accepted\n```\n\nOne prompt. 625 agents worked that problem. You got one synthesized answer.\n\n**Try it yourself:**\n```text\nhive1k \"Map this repo, explain how the major systems fit together, and list the 5 highest-risk gaps\"\n```\n\n---\n\n## Why a thousand agents?\n\nOne model gives you one perspective. For a small task, that’s fine. For anything with real stakes — architecture that spans six services, a migration that touches every API surface, a security audit where missing one edge case matters — one perspective is a gamble.\n\nThe failure that started this project: three sealed judges scored a system design 44–46 out of 50. Shadow scoring — hidden criteria the judges never saw — caught critical arithmetic errors in the same output. **Review alone is not validation.** Confident and correct are not the same thing.\n\nHive1K exists because some tasks deserve more than one brain’s best guess. It turns a single request into a structured process: decompose, parallelize across model families, cross-review, validate against sealed criteria, converge. The answer you get back isn’t one model’s opinion — it’s the output that survived a gauntlet.\n\nAnd it scales sub-linearly. Five times more agents costs roughly 2.2× more wall-clock time. The architecture is parallelism-first, convergence-second.\n\n---\n\n## The Hive\n\nPicture it as a living organization, not a diagram.\n\n**Nexus** sits at the top — one orchestrator running on `claude-opus-4.6` with a 128K context budget. It reads your mission, decides what divisions are needed, and seals the acceptance criteria in an envelope that no agent below will ever see.\n\n**Division Commanders** — up to four, named DIV-ALPHA through DIV-DELTA — each own a slice of the mission. They’re the recursive layer that Hive1K’s predecessor ([Swarm Command](https://github.com/DUBSOpenHub/swarm-command)) didn’t have. This is what lets the system scale from hundreds to over a thousand agents without losing coherence.\n\n**Commanders** (20 total, 5 per division) are domain specialists — architecture, implementation, testing, documentation, integration. They break their domain into micro-tasks.\n\n**Squad Leads** (200 total) decompose those micro-tasks further and run canary checks before committing workers.\n\n**Workers** (1,000 total) do the actual atomic work. They’re leaf nodes — they execute, they don’t spawn. Each gets a 128-token micro-brief and returns a 256-token atom.\n\n**Reviewers** (20 total) form a cross-family mesh. Every review pair is intentionally split across model families — Claude reviews GPT’s work, GPT reviews Claude’s — so agreement means more than self-consistency.\n\nThen the sealed envelope opens. **Shadow scoring** validates everything against criteria the swarm never optimized for. If the score is too high, a hardening cycle fires. Only what survives gets synthesized into your final answer.\n\n```\n                            Your question\n                                 │\n                                 ▼\n                    ┌────────────────────────┐\n                    │       NEXUS (1)        │  Decomposes mission\n                    │  Seals hidden criteria │  Synthesizes final answer\n                    └───────────┬────────────┘\n                                │\n            ┌───────────┬───────┴───────┬───────────┐\n            ▼           ▼               ▼           ▼\n        DIV-ALPHA   DIV-BETA      DIV-GAMMA   DIV-DELTA    4 Division\n        ┌─┴─┐       ┌─┴─┐         ┌─┴─┐       ┌─┴─┐       Commanders\n        │   │       │   │         │   │       │   │\n       ─┴─ ─┴─    ─┴─ ─┴─      ─┴─ ─┴─    ─┴─ ─┴─        20 Commanders\n       │││ │││    │││ │││      │││ │││    │││ │││          (5 per division)\n       ··· ···    ··· ···      ··· ···    ··· ···\n       │││ │││    │││ │││      │││ │││    │││ │││          200 Squad Leads\n       ▼▼▼ ▼▼▼    ▼▼▼ ▼▼▼      ▼▼▼ ▼▼▼    ▼▼▼ ▼▼▼          (10 per commander)\n      ┌──────────────────────────────────────────────┐\n      │         1,000 Workers execute in parallel     │\n      │     Atomic tasks · 8K context · Leaf nodes    │\n      └──────────────────────┬───────────────────────┘\n                             │\n                             ▼\n      ┌──────────────────────────────────────────────┐\n      │    20 Reviewers — cross-family scoring mesh   │\n      │  Claude ↔ GPT pairs · 4-axis sealed scoring  │\n      └──────────────────────┬───────────────────────┘\n                             │\n                             ▼\n                  ┌──────────────────────┐\n                  │   Shadow Scoring     │  Sealed-envelope validation\n                  │   Hardening cycle    │  Spec L2 conformance\n                  └──────────┬───────────┘\n                             │\n                             ▼\n                        Your answer\n```\n\nContext compresses on the way down. Results compress on the way up. A 4K-token mission becomes 128-token micro-briefs at the leaves, and 256-token atoms bubble back up through merges until Nexus holds a 4K-token final report. Nothing explodes.\n\nFor the full visual deep dive: [architecture diagrams](docs/architecture-diagrams.md) · [architecture overview](docs/architecture.md)\n\n---\n\n## Scale\n\nHive1K isn’t one size. It’s three sizes, and you pick the one that fits.\n\n### H-250 · Scout Swarm\n\n**~316 agents** · Fast reconnaissance. Bounded tasks where you want multi-model coverage without the full hierarchy. Good for: mapping a codebase, reviewing a design doc, triaging a bug backlog.\n\n### H-500 · Worker Swarm\n\n**~625 agents** · The workhorse. Most real software tasks land here — document a system, write missing tests, audit for security gaps, plan a migration. Two Division Commanders coordinate the effort.\n\n### H-1K · Full Hive\n\n**~1,245 agents** · Maximum coverage. Four Division Commanders, the complete hierarchy, every model family engaged. For repo-wide audits, high-stakes architecture decisions, or anything where missing a blind spot has real consequences.\n\n```text\nhive1k h-250 \"Triage the open bug backlog and rank by risk\"\nhive1k h-500 \"Document the auth system and flag rollout risks\"\nhive1k h-1k  \"Full architecture review — find every gap, test every assumption\"\n```\n\nCost scales sub-linearly (α ≈ 0.45). Going from H-250 to H-1K is roughly 2.2× the wall-clock time, not 4×. The architecture pays for parallelism, not for waiting.\n\nDetails and cost estimates: [docs/scaling.md](docs/scaling.md)\n\n---\n\n## Trust\n\nA thousand agents can produce a lot of confident nonsense if you let them. Hive1K is built around the premise that **agreement is not accuracy** — three agents saying the same wrong thing is worse than one agent saying it, because it feels more true.\n\n### Consensus scoring\n\nA 4-stage pipeline decides what survives:\n\n1. **Workers self-score** — each atom ships with a confidence signal\n2. **Squad Leads merge locally** — classify results as CONSENSUS / MAJORITY / CONFLICT / UNIQUE\n3. **Commanders merge across squads** — trimmed mean, weighted formula:\n   `0.40 × confidence + 0.30 × evidence + 0.15 × scope + 0.15 × coverage − conflict_penalty`\n4. **Nexus arbitrates** — median-of-3 judging on unresolved conflicts\n\nDisagreement isn’t suppressed. It’s scored, preserved, and surfaced. When agents conflict, you see it.\n\n### Shadow scoring\n\nThe sealed-envelope protocol. Before any commander executes, Nexus generates acceptance criteria and locks them away. The swarm never sees them. After execution, the criteria unseal and validate the output.\n\n| Shadow Score | What it means | What happens |\n|---|---|---|\n| 0% | Every criterion passed | Ship it |\n| 1–15% | Minor gaps | Proceed with notes |\n| 16–30% | Moderate gaps | Gap report attached, warning raised |\n| 31–50% | Significant gaps | Bundle quarantined, hardening cycle |\n| \u003e 50% | Critical failure | Bundle rejected entirely |\n\nThis is why Hive1K exists as a distinct project. Its predecessor didn’t have this. The moment we saw judges rate flawed output 44/50 while hidden criteria caught the errors, shadow scoring became non-negotiable.\n\nFull protocol: [docs/shadow-scoring.md](docs/shadow-scoring.md) · [docs/consensus.md](docs/consensus.md)\n\n---\n\n## Get started\n\n### Quickstart (one command)\n\n```bash\ncurl -fsSL https://raw.githubusercontent.com/DUBSOpenHub/hive1k/main/quickstart.sh | bash\n```\n\nThen open Copilot and type: `hive1k`\n\nPrefer to inspect first?\n```bash\ncurl -fsSL https://raw.githubusercontent.com/DUBSOpenHub/hive1k/main/quickstart.sh -o quickstart.sh\nless quickstart.sh\nbash quickstart.sh\n```\n\n### Manual install\n\n```bash\nmkdir -p ~/.copilot/skills/hive1k ~/.copilot/agents \u0026\u0026 \\\n  curl -sL https://raw.githubusercontent.com/DUBSOpenHub/hive1k/main/skills/hive1k/SKILL.md \\\n    -o ~/.copilot/skills/hive1k/SKILL.md \u0026\u0026 \\\n  curl -sL https://raw.githubusercontent.com/DUBSOpenHub/hive1k/main/agents/hive1k.agent.md \\\n    -o ~/.copilot/agents/hive1k.agent.md\n```\n\n### Verify your downloads\n\n```bash\n# macOS\nshasum -a 256 ~/.copilot/skills/hive1k/SKILL.md\nshasum -a 256 ~/.copilot/agents/hive1k.agent.md\n\n# Linux\nsha256sum ~/.copilot/skills/hive1k/SKILL.md\nsha256sum ~/.copilot/agents/hive1k.agent.md\n```\n\n\u003e **Note:** SHA hashes are published in the [latest release](https://github.com/DUBSOpenHub/hive1k/releases/latest). Compare your download hashes against the release notes before use.\n\n### Clone the repo\n\n```bash\ngit clone https://github.com/DUBSOpenHub/hive1k.git\ncd hive1k\nchmod +x quickstart.sh \u0026\u0026 ./quickstart.sh\n```\n\n*Requires an active [Copilot subscription](https://github.com/features/copilot/plans).*\n\n---\n\n## Neighbors, not competitors\n\nHive1K is one tool in a family. They solve different problems.\n\n| You want to... | Use | Why |\n|---|---|---|\n| Get **one consensus answer** from a recursive agent hierarchy (250–1,245 agents) | [**Hive1K**](https://github.com/DUBSOpenHub/hive1k) | Recursive decomposition, cross-model review, shadow validation, one synthesized output |\n| Run **parallel coding workstreams** across terminals | [**Stampede**](https://github.com/DUBSOpenHub/terminal-stampede) | Independent task lanes, execution throughput, branch-per-task |\n| **Tournament-test ideas** across many models | [**Havoc Hackathon**](https://github.com/DUBSOpenHub/havoc-hackathon) | Competitive elimination rounds, sealed judge panels, ranked synthesis |\n| Orchestrate **~250 agents** without the recursive layer | [**Swarm Command**](https://github.com/DUBSOpenHub/swarm-command) | Hive1K’s predecessor — same core, no Division Commanders, depth 3 instead of 4 |\n\n**Short version:** Hive1K for consensus at scale. Stampede for parallel execution. Havoc for idea tournaments. Swarm Command if you want the simpler original.\n\n---\n\n## Under the hood\n\nThe sections below are for people who want to know how the gears turn. If you just want to use Hive1K, everything above is enough.\n\n### The 16 models\n\n| Role | Models |\n|---|---|\n| **Nexus** | claude-opus-4.6 |\n| **Commanders** (pool of 10) | claude-opus-4.6, claude-opus-4.5, claude-opus-4.6-1m, claude-sonnet-4.6, claude-sonnet-4.5, claude-sonnet-4, gpt-5.4, gpt-5.2, gpt-5.1, goldeneye |\n| **Squad Leads** | claude-haiku-4.5, gpt-5.4-mini |\n| **Workers** (pool of 6) | claude-haiku-4.5, gpt-5.4-mini, gpt-5-mini, gpt-4.1, gpt-5.3-codex, gpt-5.2-codex |\n| **Reviewers** (8 cross-family pairs) | claude-opus-4.6↔gpt-5.4, claude-opus-4.5↔gpt-5.2, claude-opus-4.6-1m↔gpt-5.1, claude-sonnet-4.6↔gpt-5.3-codex, claude-sonnet-4.5↔gpt-5.2-codex, claude-sonnet-4↔gpt-5.4-mini, claude-haiku-4.5↔gpt-5-mini, goldeneye↔gpt-4.1 |\n\nEvery reviewer pair intentionally crosses model families. When Claude and GPT agree, that signal is worth more than either alone.\n\n### Configuration\n\nAll tunables live in [`config.yml`](config.yml):\n\n```yaml\nconsensus:\n  threshold_consensus: 0.70\n  threshold_majority: 0.50\n\ndepth_guard:\n  max_spawn_depth: 4\n  max_workers_per_squad_lead: 5\n\ncircuit_breaker:\n  timeout_cascade: [240, 150, 90, 50, 30]\n\nshadow_scoring:\n  enabled: true\n  spec_version: \"1.0.0\"\n  conformance_level: \"L2\"\n  sealed_criteria_count: 10\n  hardening:\n    enabled: true\n    threshold: 15\n\ncost_ceiling:\n  enabled: true\n  mode: user-configurable\n```\n\n**Depth Guard** enforces 5 laws and 3 layers of protection against runaway recursion. **Circuit breaker** implements a 3-state FSM with 5-level recovery escalation. Neither is optional — at this scale, guardrails are structural.\n\n### Safety mechanisms\n\n- **Depth Guard** — hard limit on recursion depth (max 4), spawn budgets per layer, enforcement at every level\n- **Circuit breaker** — CLOSED → OPEN → HALF-OPEN FSM with cascading timeouts\n- **Token compression** — context shrinks at each layer (4K → 3K → 2K → 512 → 128 tokens), results compress on the way back up\n- **Cost ceiling** — user-configurable budget cap; the swarm stops before it overspends\n\n### Repo structure\n\n```text\nhive1k/\n├── README.md                           # You are here\n├── AGENTS.md                           # Agent/skill descriptions\n├── CONTRIBUTING.md                     # Contribution guidelines\n├── catalog.yml                         # Skill metadata\n├── config.yml                          # All tunables\n├── LICENSE                             # MIT\n├── SECURITY.md                         # Security policy\n├── quickstart.sh                       # One-line installer\n├── .github/\n│   ├── copilot-instructions.md         # AI agent instructions for this repo\n│   ├── workflows/ci.yml                # CI validation\n│   └── skills/hive1k/SKILL.md         # Skill discovery path\n├── agents/\n│   └── hive1k.agent.md                # Standalone agent version\n├── skills/hive1k/\n│   └── SKILL.md                        # Core skill definition\n├── templates/\n│   ├── commander.md                    # Commander prompt template\n│   ├── division-commander.md           # Division Commander prompt template\n│   ├── worker.md                       # Worker prompt template\n│   ├── reviewer.md                     # Cross-reviewer prompt template\n│   └── squad-lead.md                   # Squad lead prompt template\n├── protocols/\n│   ├── depth-guard.md                  # 5 Laws + 3-layer enforcement\n│   ├── circuit-breaker.md              # 3-state FSM + 5-level recovery\n│   ├── context-capsule.md              # JSON schemas for data structures\n│   └── meta-reviewer.md               # Reviewer quality gate protocol\n└── docs/\n    ├── architecture.md                 # Architecture overview\n    ├── architecture-diagrams.md        # Mermaid diagrams\n    ├── consensus.md                    # Consensus algorithm deep dive\n    ├── example-output.md               # Sample completed run output\n    ├── learning-path.md                # Recommended reading order\n    ├── scaling.md                      # Scale chooser + cost estimates\n    ├── shadow-scoring.md               # Shadow scoring protocol\n    └── use-cases.md                    # Expanded prompt gallery\n```\n\n### Go deeper\n\n| Doc | What’s in it |\n|---|---|\n| [learning-path.md](docs/learning-path.md) | Beginner, operator, and architect reading tracks |\n| [architecture.md](docs/architecture.md) | The full system model |\n| [architecture-diagrams.md](docs/architecture-diagrams.md) | Mermaid diagrams for every layer |\n| [scaling.md](docs/scaling.md) | Scale chooser, cost estimates, tuning guide |\n| [use-cases.md](docs/use-cases.md) | Prompt gallery with expected outcomes |\n| [consensus.md](docs/consensus.md) | The 4-stage consensus algorithm in detail |\n| [shadow-scoring.md](docs/shadow-scoring.md) | The sealed-envelope protocol, hardening cycle |\n| [example-output.md](docs/example-output.md) | Full transcript of a completed swarm run |\n\n---\n\n## Spec conformance\n\nHive1K implements **[Shadow Score Spec](https://github.com/DUBSOpenHub/shadow-score-spec) L2** — sealed acceptance criteria generated before execution, validated after, hardened on failure.\n\n---\n\n## License\n\n[MIT](LICENSE) — use it, fork it, build on it.\n\n---\n\n\u003cp align=\"center\"\u003e\n🐝 Built by \u003ca href=\"https://github.com/DUBSOpenHub\"\u003e@DUBSOpenHub\u003c/a\u003e with the \u003ca href=\"https://docs.github.com/copilot\"\u003eGitHub Copilot CLI\u003c/a\u003e\n\u003c/p\u003e\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fdubsopenhub%2Fhive1k","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fdubsopenhub%2Fhive1k","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fdubsopenhub%2Fhive1k/lists"}