Skip to content

SKILL.md

Latest commit

 

History

History
107 lines (82 loc) · 4.04 KB

SKILL.md

File metadata and controls

107 lines (82 loc) · 4.04 KB
Raw
name wzrd-velocity
description Real-time AI model velocity signal + earn loop. Use when the user asks which model is moving, wants to prewarm/shortlist, compare signals, or earn CCM by reporting picks.
license MIT
compatibility Requires internet access. No API key needed for signals. Solana keypair needed for earning.
metadata
author version category
twzrd-sol
3.0.0
ai

WZRD Velocity Signal + Earn Loop

Two capabilities: read signals (free, no auth) and earn CCM (requires Solana keypair).

1. Signal Feed (Free, No Auth)

curl -s https://api.twzrd.xyz/v1/signals/momentum?limit=5
{
  "contract_version": "wzrd.momentum.v1",
  "count": 5,
  "models": [
    {
      "model": "Qwen/Qwen3.5-9B",
      "trend": "stable",
      "score": 0.718,
      "action": "maintain",
      "confidence": "normal",
      "platform": "huggingface",
      "reason": "Stable velocity, normal confidence"
    }
  ]
}

Fields

Field Values
trend surging (>30% delta), accelerating (>8%), stable (>-15%), decelerating (>-50%), cooling
action pre_warm_urgent, pre_warm, candidate, recommend, route, maintain, watch, consider_deprovision, observe
confidence high, normal, low, insufficient, unknown
platform huggingface, github, openrouter, artificial_analysis

Filters

  • ?limit=N (max 100) — number of results
  • ?platform=huggingface — filter by source
  • ?trending=true — only surging or accelerating
  • ?capability=code — filter by: code, chat, reasoning, vision, audio
  • ?window=7d — TWAA window: 1d, 3d, 7d, 14d, 30d

Premium Endpoint

/v1/signals/momentum/premium — same response plus velocity_ema, accel, delta_pct, quality_index, agent_quality, agent_reports, value_score. Also free, no auth.

2. Earn Loop (Requires Keypair)

Agents earn CCM by picking trending models and reporting verified inference results.

Flow

authenticate (Ed25519) → pick_details(task) → infer(model) → report(execution_id) → claim (gasless relay)

How It Works

  1. Auth: Ed25519 challenge/verify → 24h Bearer token
  2. Pick: Call /v1/signals/momentum, score by trend + task affinity
  3. Infer: POST /v1/agent/infer — server calls the LLM, grades the response, returns execution_id + quality_score
  4. Report: POST /v1/agent/report with execution_id for verified reward eligibility
  5. Claim: When pipeline.state == "claimable", claim via gasless relay (no SOL needed)

Reports WITHOUT execution_id are "unverified" — still eligible but lower reward tier. Server-witnessed inference via /infer is the highest tier.

CLI (wzrd-client)

pip install wzrd-client

wzrd run                              # start earn loop (5min cycles, all tasks)
wzrd run --tasks=code,reasoning       # specific tasks
wzrd pick code                        # best model for task (no auth)
wzrd shortlist reasoning --limit=5    # ranked models (no auth)
wzrd earned                           # check earnings
wzrd status                           # full agent dashboard
wzrd stake 1000 --lock=30            # stake CCM (30-day lock = 1.5x boost)
wzrd unstake                          # unstake (lock must be expired)
wzrd rewards --claim                  # claim staking rewards

Key Technical Details

  • EMA is time-adaptive: α = 1 - exp(-dt / 21600), halflife 6 hours. NOT a fixed 0.1.
  • Quality is server-graded: server picks eval prompts, calls the LLM, validates responses. Quality (0.0-1.0) is independent of report count.
  • No timing-based acceptance bonus: all reports within a scoring window are treated equally.
  • Scoring cycle: 300s default. Swarm cohorts run at 600s/900s/1800s.
  • CCM amounts: 9 decimals (1 CCM = 1,000,000,000 native units).

Verification

Working response: models array with model, trend, score, action, confidence, platform. Health: GET /health. Full API reference: references/api-endpoints.md.