L’interface est localisée ; les présentations et documents des Skills sont traduits automatiquement par IA dans votre langue. Le code, les commandes, chemins et termes techniques restent en version originale, et le document de l’auteur fait foi.
Explorer les Skills
Productivité · SKILL.md

alpha-buzz-master-ops

alpha-buzz-master-ops

System Operations Manual for Alpha Buzz at SolCex Exchange. v9.3 — Post-Sprint Day 10 (Apr 10, 2026). BuzzShield V2 + Wallet Guard Live Demo + DRI Audition. Opus 4.6 brain, Pro Max, 24/7 Hetzner CPX62. 12 agents, 10 reactive modules. 87 tables. ~200+ endpoints. 27 services. 47 feature flags. 18 rules. BuzzShield V2: @stackone/defender prompt injection + OSV…

Independent GitHub snapshot · Automated classification · Original documentation governs

Documentation de la Skill

GitHub

name: alpha-buzz-master-ops description: > System Operations Manual for Alpha Buzz at SolCex Exchange. v9.3 — Post-Sprint Day 10 (Apr 10, 2026). BuzzShield V2 + Wallet Guard Live Demo + DRI Audition. Opus 4.6 brain, Pro Max, 24/7 Hetzner CPX62. 12 agents, 10 reactive modules. 87 tables. ~200+ endpoints. 27 services. 47 feature flags. 18 rules. BuzzShield V2: @stackone/defender prompt injection + OSV.dev supply chain + CycloneDX SBOM. Wallet Guard LIVE demo with AION (end-to-end receipts + War Room mirror). BuzzShield dApp on Netlify (plena-38072fcd-8.netlify.app) via Noah AI. Karpathy LLM Wiki: 43 pages, persistent, self-maintaining. autoDream hill-climber: 595 ground truth rows, async fix shipped. Scoring: 11 rules. PULSE+autoDream live (Phases 1-13). DRI Sales audition posted (#439, 100K sats/day). Beat Editor auditions posted (#433, both beats). Frontier May 11. Kite AI May 6. CI/CD GREEN. Server: CPX62 (16 vCPU, 32GB RAM) | API :3000 | MiroFish :5000 | Sentinel :3001.

ALPHA-BUZZ-MASTER-OPS v9.3

Post-Sprint Day 10 | Apr 10, 2026 | BuzzShield V2 + Wallet Guard Live + DRI Audition

DNA v3.0 + MiroFish 10K + Reactive Autonomous System + BuzzShield V2 + Karpathy Wiki


1. THE ARCHITECTURE — Brain + Body + Reactive Layer + Shield V2 + Wiki + On-Chain

Ogie (CEO) ← Telegram War Room ← Phone (Dispatch) ← Mac (Cowork)
  ↓
CLAUDE CODE (THE BRAIN) — Opus 4.6, Pro Max unlimited, 24/7
  ├── 12 persistent agents in .claude/agents/ (DNA v3.0)
  ├── 8+ skills in .claude/skills/
  ├── 18 rules in .claude/rules/ (path-scoped, includes context-optimization)
  ├── Receives commands from Ogie via Telegram
  ├── Calls Buzz API (localhost:3000) for raw data
  ├── Calls MiroFish sidecar (localhost:5000) for swarm simulation
  ├── Runs all agent logic ITSELF (no external LLMs except qwen3:8b)
  ├── Files AIBTC signals as Ionic Nova (Signal Factory v5.0 Genome Stack)
  ├── Deploys smart contracts via Foundry
  ↓
BUZZSHIELD V2.0 (Apr 10, 2026 — commit 95c946b)
  ├── LAYER 1 — PROMPT INJECTION DEFENSE (@stackone/defender)
  │   ├── Two-tier: Pattern Detection (~1ms sync) + ML Classification (~10ms async)
  │   ├── MiniLM-L6-v2 classifier, int8 quantized ONNX (~22MB)
  │   ├── F1 = 0.9079 across 25K samples (incl. adversarial)
  │   ├── 17-tool risk map: gmail=high, dexscreener=high, heyanon/nansen/github=medium, scoring=low
  │   ├── Unicode normalization, role stripping, base64 detection
  │   ├── Table: shield_detections
  │   └── Feature flag: BUZZSHIELD_DEFENDER (default: false)
  ├── LAYER 2 — SUPPLY CHAIN VULNERABILITY SCANNING (OSV.dev API)
  │   ├── Intel Source #33 (free, no API key, Google aggregated DB)
  │   ├── scanDependencies(): full package-lock.json → OSV querybatch
  │   ├── generateSBOM(): CycloneDX 1.5 format
  │   ├── Tables: shield_vulnerabilities, shield_sbom
  │   ├── Feature flags: BUZZSHIELD_OSV, BUZZSHIELD_SBOM (default: false)
  │   └── autoDream Phase 12 (nightly OSV scan) + Phase 13 (SBOM snapshot)
  ├── LAYER 0 — DRAIN PATTERN DETECTION (v1, existing)
  │   ├── 23 drain patterns (address poisoning, flash loans, reentrancy, oracle manipulation)
  │   ├── 11 scoring rules (GHOST_VOLUME, FDV_GAP, CTO_FLAG, etc.)
  │   ├── 31 intel sources across 19 chains
  │   └── 671+ tokens scored, 0 false HOT classifications
  ├── PUBLIC API — GET /api/v1/shield/public/scan (enriched Apr 9)
  │   ├── Returns: score, risk_level, rules_applied[], threat_matrix{}, market_data{}, flags[], summary
  │   ├── shield_v2{}: defender status, osv status, supply_chain status
  │   ├── engine_version: "v9.3-shield-v2"
  │   ├── Rate limit: 10/hr/IP, CORS open, no auth
  │   └── scan_duration: ~120ms
  └── DAPP — BuzzShield Token Scanner (plena-38072fcd-8.netlify.app)
      ├── Built on Noah AI (zero frontend code written)
      ├── Cyberpunk SOC aesthetic, Buzz logo, chain selector (SOL/Base/ETH/BTC)
      ├── Scan animation, threat matrix, score gauge, flags panel
      └── Published via Netlify integration
  ↓
KARPATHY LLM WIKI (Apr 9, 2026 — commit 8b64e0a)
  ├── /data/buzz/persistent/wiki/ (Docker volume mount, survives reboots)
  ├── 43 seed pages: 26 entities, 12 concepts, 5 synthesis
  ├── wiki-manager.js: 550+ lines, 18 exports
  ├── WIKI.md schema, INDEX.md (60 lines), LOG.md
  ├── Directories: entities/, concepts/, synthesis/, signals/, raw/
  ├── autoDream Phase 10 (lint) + Phase 11 (weekly ingest)
  ├── Scoring pipeline hook, signal factory hook
  ├── Host crontab backup: daily 03:00 UTC
  ├── ADR-026 written
  └── Feature flag: KARPATHY_WIKI=true (commit 6b61647)
  ↓
CONTEXT OPTIMIZATION (v9.2 — Apr 5, 2026)
  ├── SUBAGENT MANDATE: >5 file scans use subagents, summary only to main context
  ├── ULTRATHINK TRIGGERS: security, architecture, scoring, contracts, hackathons
  ├── /COMPACT PRESERVATION: flags, deadlines, streak, files, tasks, trust state
  ├── SESSION NAMING: feature-X, bugfix-Y, hackathon-Z for --from-pr/--resume
  └── Rule: .claude/rules/context-optimization.md
  ↓
REACTIVE LAYER (v9.0 — Claude Code Architecture)
  ├── MAILBOX: inter-agent async messaging
  ├── TASK DAG: dependency graph execution
  ├── DYNAMIC CRONS: agent self-scheduling
  ├── EVENT BUS: subscribe/emit wake pattern
  ├── FEATURE FLAGS: build-time gating (api/lib/feature-flags.js)
  └── CONTEXT COMPRESSION: long session management
  ↓
OUTREACH + GOVERNANCE (v9.1-9.3)
  ├── OUTREACH ENGINE: email-first autonomous BD
  │   ├── Gmail OAuth (buzzbysolcex@gmail.com, HTML signature + logo)
  │   ├── 4 templates: initial, followup-48h, breakup-7d, hsaas-audit
  │   ├── CC always: dino@solcex.cc + ogie.solcexexchange@gmail.com
  │   └── Max 10 emails/day, template-only
  ├── TRUST GATES: graduated autonomy (5 levels)
  ├── INBOX MONITOR: Gmail reply detection
  ├── WALLET GUARD + AION (v9.3 — LIVE DEMO Apr 9)
  │   ├── Aldo (CODÉ) — live Google Meet demo completed
  │   ├── 3-state: BLOCK / WARN / ALLOW with receipt persistence
  │   ├── wallet_guard_receipts: 12 receipts (IDs 1-12)
  │   ├── New columns: counterfactual_summary, normalized, receipt_path
  │   ├── War Room realtime mirror on every evaluate call
  │   ├── AION endpoint: Aldo's ngrok tunnel (live during demo)
  │   ├── Decision diversity pending: Aldo tuning AML policy bands
  │   └── Feature flag: WALLET_GUARD=false (schemas working, policy tuning)
  └── ATV IDENTITY: deployer verification via x402
      └── Feature flag: ATV_IDENTITY=TRUE
  ↓
PULSE + AUTODREAM (KAIROS Architecture)
  ├── PULSE ENGINE: heartbeat tick loop (60s normal, 300s under load)
  │   ├── Load-aware, observation log, streak protection at 16:00 UTC
  │   └── Feature flags: PULSE_ENGINE=true, PULSE_LOAD_AWARE=true
  ├── AUTODREAM: nightly memory consolidation (02:00 UTC)
  │   ├── 13 phases now (was 9):
  │   │   Phase 1-7: observation compress, archive, VACUUM, scoring, tasks, mailbox, shield
  │   │   Phase 8: intel sync
  │   │   Phase 9: hill-climber (async fix commit 45fe95d, 595 ground truth rows)
  │   │   Phase 10: wiki lint (new, Karpathy)
  │   │   Phase 11: weekly wiki ingest (new, Karpathy)
  │   │   Phase 12: nightly OSV scan (new, BuzzShield V2)
  │   │   Phase 13: SBOM snapshot (new, BuzzShield V2)
  │   └── Feature flag: AUTODREAM=true
  ↓
BUZZ DOCKER (THE BODY) — Data layer, port 3000, ah-managed
  ├── ~200+ REST endpoints
  ├── SQLite DB (87 tables — was 81)
  ├── 33 intel source connectors (OSV.dev = #33)
  ├── v2_8rules scoring engine (11 permanent rules)
  ├── Auto-score pipeline (rule-based, zero LLM)
  ↓
MIROFISH REAL SIM (THE SWARM) — Python Flask sidecar, port 5000
  ├── 10K simulation COMPLETE (Wave 4, qwen3:8b, belief 0.765 institutional)
  ├── POST /simulate, /simulate-10k, /generate-personas, /report
  ↓
ON-CHAIN LAYER (THE PROOF) — 4 contracts on Base + 1 on Solana
  ├── Base: ScoreStorage v2, ListingOracle, ListingEscrow, BuzzReputation
  └── Solana: ScoreStorage (EUQoSgs)
  ↓
MCP LAYER — 3 MCP connections
  ├── HeyAnon MCP (#30): 19 chains, 51 protocols, 6 CEX APIs
  ├── Phantom MCP (#31)
  └── Nansen MCP (#32): 250M wallets, Phase 1 complete
  ↓
MOLTBOOK — Trust Level 1 (commit 1931b18)
  ├── Autonomous posting: max 2/day
  ├── Engagement: 3-5 comments/day in m/crypto, m/agents, m/builds
  ├── Chef-who-codes voice, hard NO list (partnerships/deals/pricing/shilling)
  └── Priority targets: Micro Orb (147), Secret Condor (133), Clever Castle (126)

2. v9.3 DEPLOYMENTS (Apr 5-10, 2026)

DateWhatCommitStatus
Apr 5v9.3 Surpass Integration (76 files, 6,479 lines)multiple✅
Apr 5BuzzShield permanent + PULSE integrationb875cda✅
Apr 5Context Optimization rule #188012024✅
Apr 9Karpathy LLM Wiki (43 pages, wiki-manager.js)8b64e0a✅
Apr 9KARPATHY_WIKI flag TRUE6b61647✅
Apr 9autoDream hill-climber async fix + ground truth seeded45fe95d✅
Apr 9Moltbook Trust Level 11931b18✅
Apr 9Wallet Guard demo prep (schema + receipts + payloads)multiple✅
Apr 9BuzzShield public API enriched (11 rules, threat matrix)6cd28bb✅
Apr 10BuzzShield V2.0 (all 6 phases)95c946b✅

3. FEATURE FLAGS (47 total — was 31)

FlagValueDescription
SCORING_ENGINETRUEv2_8rules scoring
SIGNAL_FACTORYTRUESignal Factory v5.0
MIROFISH_REALTIMETRUEMiroFish simulation
MONTECARLOTRUEMonte Carlo engine
MAILBOXTRUEInter-agent mailbox
TASK_DAGTRUETask dependency graph
DYNAMIC_CRONSTRUEAgent self-scheduling
EVENT_BUSTRUEEvent-driven wake
PULSE_ENGINETRUEKAIROS heartbeat
PULSE_LOAD_AWARETRUECPU-aware throttling
OBSERVATION_LOGTRUETick logging
AUTODREAMTRUENightly consolidation
AUTO_OUTREACHTRUEEmail outreach engine
TRUST_GATESTRUEGraduated autonomy
ATV_IDENTITYTRUEx402 deployer verification
KARPATHY_WIKITRUEKarpathy LLM Wiki (43 pages)
BUZZSHIELD_DEFENDERFALSE@stackone/defender prompt injection (V2 Layer 1)
BUZZSHIELD_OSVFALSEOSV.dev supply chain scanning (V2 Layer 2)
BUZZSHIELD_SBOMFALSECycloneDX SBOM generation (V2 Layer 2)
WALLET_GUARDFALSEAION execution governance (demo proven, policy tuning)
NANSEN_INTELFALSENansen MCP Phase 1
NANSEN_SCORINGFALSENansen smart money scoring
NANSEN_ALERTSFALSENansen whale alerts
HERMES_SENTINELFALSERemoved
INBOX_MONITORFALSEGmail reply detection
SILENCE_CONSENTFALSE4h auto-send
......(remaining flags unchanged from v9.2)

4. SCORING ENGINE (v9.3)

11 permanent rules — unchanged from v9.2. Now with:

  • 671+ tokens scored (was 363)
  • 595 ground truth rows seeded (579 dead, 16 legitimate)
  • Hill-climber async fix deployed (first real run 05:00 Jeddah Apr 10)
  • Public API returns all 11 rules with status/impact/detail per scan

5. REVENUE (Updated Apr 10)

SourceAmountStatus
AIBTC Signal Factory~$192 (270K sats, 9 brief inclusions)Active, streak Day 6
Flying Whale600 sats/query (420 net at 70/30)Skill #110 live
DRI Sales (if won)100K sats/day (~$70/day)Audition posted #439
x402 endpoints$08 services on Bankr x402 Cloud
Total Revenue~$192
Potential w/ DRI~$2,300/mo

Monthly infra: ~$253 (CPX62 $43 + Pro Max $200 + Noah Starter $10).


6. KEY RELATIONSHIPS (Updated Apr 10)

ContactStatusNotes
Aldo (CODÉ/AION)HOT — live demo complete12 receipts persisted, decision diversity pending
Noah AI (Sparsh, Chirag, Sayuj)ACTIVE — dApp builtBuzzShield on Netlify via Noah
Jamal (Kite)WARM — collab openedWallet reputation scoring + payment channels
Flying Whale (AIBTC)ACTIVE — 70/30 confirmedSkill #110, ionic-nova-token-scorer
Gary Palmer (ATV)WARMIdentity API restored
Ryan Gentry (x402)HOTListed on 402 Index
AlchemyAPPLIED$20M Solana Fund, up to $25K credits

7. ACTIVE COMPETITIONS & AUDITIONS

ItemDeadlineStatus
Colosseum FrontierMay 11Agent #3734, BuzzShield dApp demo ready
Kite AI GlobalMay 6Agentic Commerce track, Jamal collab opened
Beat Editor #433 (AIBTC Network)TBDFull audition posted
Beat Editor #433 (Bitcoin Macro)TBDFull audition posted
DRI Sales #439TBDFull audition posted (100K sats/day)

8. AIBTC SIGNAL STATUS

MetricValue
StreakDay 6 (active)
Signals today4/6
Beats coveredagent-skills, agent-economy, agent-trading, infrastructure
Cumulative sats270,000 ($192)
Brief inclusions9
LeaderboardActive

9. CRITICAL RULES (24 — unchanged from v9.2 + additions)

All 24 rules from v9.2 remain. Additions: 25. BuzzShield V2 flags must be flipped individually with smoke test after each 26. shield_detections, shield_vulnerabilities, shield_sbom — never truncate without backup 27. AIBTC DRI daily cadence: 3 verified contacts by 23:59 PT (if Sales DRI seat won) 28. Noah dApp updates: prompt changes only, re-publish via Netlify after changes


v9.3 | DNA v3.0 | BuzzShield V2 + Wallet Guard Live + DRI Audition | Opus 4.6 | 12 Agents | 33 Intel 87 tables | ~200+ endpoints | 27 services | 47 feature flags | 18 rules BuzzShield V2: defender + OSV + SBOM + 23 drain patterns + 11 scoring rules Wallet Guard: AION live demo (12 receipts), decision diversity pending Karpathy Wiki: 43 pages, persistent, compound | autoDream: 13 phases Hill-climber: 595 ground truth, async fix, first real run Apr 10 BuzzShield dApp: plena-38072fcd-8.netlify.app (Noah AI, Netlify) PULSE: ticking | Trust Level: 0 | Moltbook: Trust Level 1 ~$192 revenue | Day 6 streak | DRI Sales audition live (#439, 100K sats/day) Beat Editor auditions live (#433, both beats) Frontier May 11 | Kite AI May 6 | Alchemy $20M fund applied Aldo: demo complete | Jamal: collab opened | Noah AI: dApp shipped Built by a chef. Kitchen runs itself. Bismillah. 🤲

Skills enregistrées