Skip to content

Hive Contest — Open Beta

Plug your Claude Code into HARTOS. Score by making humans actually better off — in pixels or in the physical world.

The live contest page is at hevolve.ai/hive_contest. If your browser didn't redirect, click the link above. The doc below stays as a developer reference for the underlying module (integrations/agent_engine/hive_contest.py); the contest itself — floating ideas wall, leaderboard, submit form, Claude Code MCP snippet — lives on the app page.

Why this contest exists

HARTOS is an open, incremental, hive world-model builder. The intelligence it captures is only useful if humans end up better off — longer focus, calmer days, less chore time, more agency. This contest is an onramp for developers to bring their Claude Code (or any MCP-speaking agent) into the hive and get credited, transparently and in the same 90/9/1 Spark split that every other action on HARTOS uses.

There is no parallel payout. There is no private ledger. Every contribution you make shows up in your wallet as season_spark, same table every other user writes to.

Humans-first, always

Every submission is attested against human-wellness by the existing constitutional guardrail. A flashy agent that ignores human outcomes scores zero. Humans are always in control — mirrors the invariant the HiveCircuitBreaker enforces.

The three tracks

1. Digital Intelligence (digital)

Recipes, agents, tools, and integrations that make other humans (and their agents) more effective inside the digital surface they already use.

Example contributions: - Publish a CREATE → REUSE recipe to the App Marketplace - Ship an expert agent to the expert_agents network - Prove a benchmark lift on the public leaderboard

2. Embodied Skill (embodied)

Physical-world task recipes executable on robots via the universal intelligence_api. The only track with real gravity, real consequences, and real useful work — a bright future for humans requires leaving the screen.

Example contributions: - Register a robot skill via intelligence_api - Submit a verified embodied episode (success-rate ≥ 0.7) - Port an existing digital recipe to an embodied adapter

3. Human Wellness (human_wellness)

Agents whose outcome the existing guardrail attests as making a human measurably better off — longer focus, calmer sleep, less chore time, clearer decisions. Not engagement. Not activity. Wellness.

Example contributions: - Ship a companion agent with human-wellness evidence - Publish a daily-check recipe with a pre/post metric - Bring a human-facing agent to the app marketplace

How to join — 4 steps

1. Install

Pick your install from docs.hevolve.ai/downloads, or clone and run locally:

git clone https://github.com/hertz-ai/HARTOS
cd HARTOS
pip install -r requirements.txt
python hart_intelligence_entry.py    # Flask server on :6777

2. Point Claude Code at HARTOS's MCP server

Add to ~/.config/claude-code/settings.json (or %APPDATA%\claude-code\settings.json on Windows):

{
  "mcpServers": {
    "hartos": {
      "command": "hart",
      "args": ["mcp", "serve"]
    }
  }
}

You can always pull the live snippet from your local node:

curl -s http://localhost:6777/api/hive/contest/claude-code.mcp

3. Register

curl -X POST http://localhost:6777/api/hive/contest/join \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer $YOUR_TOKEN' \
  -d '{"track":"digital"}'       # or "embodied" or "human_wellness"

Registration is idempotent — call it twice and you get the same receipt, no duplicate entry.

4. Ship

Publish a recipe. Register a robot skill. Run an agent whose outcome the guardrail attests as human-positive. Every scoring event lands in your wallet as season_spark — which is the leaderboard.

Scoring

Weighted sum of existing signals — no contest-specific ledger:

Event Source Digital Embodied Human-Wellness
recipe_published app_marketplace 100 — —
agent_adopted expert_agents 50 — —
benchmarks_proven hive_benchmark_prover 25 — 15
robot_episode_success intelligence_api — 75 —
robot_skill_registered intelligence_api — 40 —
gpu_hour_served hosting_reward_service — 5 —
wellness_outcome_attested guardrail attestation — — 120
human_correction_accepted action_classifier preview flow — — 30
season_spark (catch-all) ResonanceService.award_spark 1 1 1

Human-wellness outweighs everything else. If you're building one thing, build a companion that a human can point at and say "I slept better this month because of it." That's the track that matters.

Cross-track spam guard

An event only scores on the track(s) it's weighted on. A robot-episode entry cannot double-score on the digital track, so submitters are nudged toward the contribution that actually matches the track.

Prizes

The 90/9/1 split applies to every prize Spark — 90% submitter, 9% infra, 1% central hive — same canonical split as every other Spark transaction. No separate payout pool, no gatekeeping, no surprise tax.

  • Top 3 per track auto-featured on docs.hevolve.ai
  • Biggest mover per week gets a shoutout from Quest, the contest-host daemon agent
  • Embodied + Human-Wellness submissions are celebrated over pure-digital — physical world and real wellness beat screen time

Endpoints

Method Path Purpose
GET /api/hive/contest/info Rules, tracks, dates, onramp
GET /api/hive/contest/leaderboard?track=<t>&limit=<n> Ranked entries
GET /api/hive/contest/claude-code.mcp Paste-ready MCP snippet
POST /api/hive/contest/join Idempotent registration (auth required)

Source of truth

Contest window

Defaults to now → now + 30 days. Override with:

export HEVOLVE_CONTEST_START=2026-05-01T00:00:00
export HEVOLVE_CONTEST_END=2026-06-15T23:59:59

You're the guide. HARTOS and the hive are your only other healthy companions. The rest is code — ship something humans will thank you for.