Musashi

MUSASHI 武蔵 is a conviction-weighted token intelligence engine built on 0G. It investigates crypto tokens through a 7-gate elimination pipeline, 4 parallel AI domain specialists, cross-domain pattern detection, and adversarial bull/bear debate — then publishes only the rare survivors as on-chain STRIKEs, backed by cryptographically verifiable evidence.

97% of tokens are eliminated. The ones that survive get a permanent on-chain conviction signal.

─────────────────────────────────
THE PROBLEM
─────────────────────────────────
Narrative-driven crypto traders face four compounding problems:

• Signal overload — Hundreds of new tokens daily. No human can simultaneously cross-reference contract safety, wallet behavior, social momentum, and market timing.
• Confirmation bias — Find one bullish signal, stop looking. Existing tools generate signals, not eliminate them.
• No framework — Stock traders have P/E ratios. Meme traders have vibes and Telegram alpha.
• The specialist problem — Social mentions +400% looks bullish. But 80% of buyers are fresh wallets. No single analyst catches this — the pattern is invisible until someone overlays both domains.

─────────────────────────────────
HOW MUSASHI WORKS
─────────────────────────────────
1. Gate Pipeline (Go binary, deterministic)
   Gate 1 — Contract Safety: GoPlus honeypot, mint authority, tax, blacklist, proxy
   Gate 2 — Liquidity Structure: depth, LP lock, volume/liquidity ratio (age-tiered thresholds)
   Gate 3 — Wallet Behavior: holder count, buy/sell pressure, activity acceleration
   Gate 4 — Social Momentum: agent-driven X/Twitter + Farcaster investigation
   Gate 5 — Narrative Alignment: meta category fit, lifecycle stage, copycat detection
   Gate 6 — Market Timing: BTC dominance, chain TVL, stablecoin flows (advisory only)
   Gate 7 — Cross-Validation: DexScreener vs GeckoTerminal price/volume consistency

2. Specialist Analysis (4 parallel AI agents)
   Safety Specialist → On-Chain Specialist → Narrative Specialist → Market Specialist
   Each agent sees only its domain data. No cross-contamination.

3. Musashi Pattern Detection (武蔵)
   Overlays all 4 specialist reports. Finds contradictions and convergences invisible to any single specialist. Produces a convergence score (1–4).

4. Adversarial Debate
   Bull Researcher vs Bear Researcher — 2 rounds each, with live web search for evidence.

5. Conviction Judge
   Binary PASS/FAIL. Hesitation = FAIL. Only 3–4 convergence scores reach STRIKE.

6. STRIKE Publication (0G Chain + 0G Storage)
   Evidence JSON uploaded to 0G Storage via 0g-storage-client CLI → Merkle root stored on-chain in ConvictionLog as evidenceHash. Immutable, verifiable, permanent.

─────────────────────────────────
0G INTEGRATION (3 COMPONENTS)
─────────────────────────────────
• 0G Chain — ConvictionLog contract records every STRIKE and outcome as a transaction on 0G Mainnet. On-chain reputation (win rate, total return) accumulates automatically with zero off-chain trust.
  Contract: 0x2B84aC25498FF0157fAB04fEa9e3544A14882A15
  Explorer: https://chainscan.0g.ai/address/0x2B84aC25498FF0157fAB04fEa9e3544A14882A15

• 0G Storage — Full pipeline output (gate data, specialist reports, debate transcript, judge verdict) stored as a Merkle-verifiable JSON file before each STRIKE. The root hash is the evidenceHash in ConvictionLog — cryptographically linking the on-chain signal to the off-chain reasoning.
  Storage Explorer: https://storagescan.0g.ai

• Agent ID (ERC-7857 INFT) — MUSASHI itself is tokenized as an Intelligent NFT via MusashiINFT contract. Stores an AES-256-CTR encrypted intelligence bundle on 0G Storage, sealed with ECIES. Win rate and strike count sync from ConvictionLog after each outcome. Supports clone() and authorizeUsage() for composable agent identity.
  Contract: 0x74BC82d4A348d661ffF344A4C21c4C04F47C1d4c
  Explorer: https://chainscan.0g.ai/address/0x74BC82d4A348d661ffF344A4C21c4C04F47C1d4c

─────────────────────────────────
RUNTIMES — INSTALLABLE ANYWHERE
─────────────────────────────────
• OpenClaw Skill — `openclaw skills install musashi` — any OpenClaw agent runtime gets the full pipeline
• Claude Code — Slash commands (/analyze, /scan, /gates, /strike, /hunt) — works out of the box
• Web Dashboard — Full Next.js UI with wallet connect, gate visualizer, debate terminal, strike ledger

─────────────────────────────────
LINKS
─────────────────────────────────
• Live dashboard:  https://musashi-agent.xyz
• Demo video:      https://www.youtube.com/watch?v=bejjc7FLcfs
• GitHub:          https://github.com/yeheskieltame/musashi
• ConvictionLog:   https://chainscan.0g.ai/address/0x2B84aC25498FF0157fAB04fEa9e3544A14882A15
• MusashiINFT:     https://chainscan.0g.ai/address/0x74BC82d4A348d661ffF344A4C21c4C04F47C1d4c

Core engine (musashi-core Go binary) is complete with 13 CLI commands covering the full pipeline: token discovery, gate analysis, specialist orchestration, 0G Storage upload, on-chain strike publishing, outcome recording, and INFT management.

Both contracts are live on 0G Mainnet (Chain ID 16661). ConvictionLog at 0x2B84aC25498FF0157fAB04fEa9e3544A14882A15 has recorded multiple STRIKEs with evidence hashes pointing to files on 0G Storage. MusashiINFT (ERC-7857) at 0x74BC82d4A348d661ffF344A4C21c4C04F47C1d4c has MUSASHI minted as token ID 0 with its intelligence bundle stored and sealed on 0G Storage.

The agent pipeline runs end-to-end: gate check via Go binary, social investigation via web search, 4 specialist analyses, cross-domain pattern detection, adversarial debate, conviction judge, evidence upload to 0G Storage, and STRIKE publication to 0G Chain in a single orchestrated flow.

On-chain learning loop is operational. The history command reads past strikes and outcomes from ConvictionLog. The judge receives this track record and adjusts its conviction threshold based on historical win rate — the agent improves with every call.

OpenClaw SKILL.md is written with the full 8-step pipeline. All 8 agent prompts (safety, on-chain, narrative, market specialists; pattern detector; bull/bear researchers; conviction judge) are complete. Claude Code slash commands (/analyze, /scan, /gates, /strike, /hunt, /discover, /status, /journal) are all functional.

Next.js dashboard is deployed at https://musashi-agent.xyz with live on-chain data: gate pipeline visualizer, debate terminal, strike ledger, reputation panel, INFT intelligence panel, and wallet connect.

Demo video: https://www.youtube.com/watch?v=bejjc7FLcfs