Blockchain web3 orchestrator

Blockchain / Web3 Orchestrator

The single entry skill for EVM and prediction-market work. It places the task on the surface × concern map — what is touching value (a contract, an autonomous agent, a market signal) crossed with which failure mode (a silent correctness bug, an exploit, or an unreviewed risk) — and delegates to one of six specialist spokes. The cross-cutting model every spoke shares — the trust boundary, the value-at-risk hierarchy, and the units/hashing conventions that make on-chain math correct — lives in blockchain-web3-core; read it before auditing a contract or wiring an agent to a wallet.

Routing map (intent → spoke)

Smart-contract correctness & exploits

• Auditing or writing a Solidity AMM / LP vault / swap flow → defi-amm-security (reentrancy, CEI, donation/inflation, oracle manipulation, slippage, admin gating)
• Reading balances / pricing / cross-chain amounts that could silently mis-scale → evm-token-decimals (runtime decimals(), chain-aware caching, bridged drift)

Off-chain integration bugs (JS/TS)

• Hashing Ethereum data, selectors, EIP-712, storage slots, addresses in Node → nodejs-keccak256 (Keccak-256 vs NIST SHA3 — the silent mismatch)

Autonomous execution & agents

• An LLM/agent that signs or sends transactions, places orders, or touches a treasury → llm-trading-agent-security (prompt injection as a financial attack, spend limits, pre-send simulation, circuit breakers, MEV, key isolation)

Prediction markets as signal

• Using market-implied probabilities as a data/oracle/decision-intelligence input → prediction-market-oracle-research (signal quality, liquidity, resolution authority, integration patterns)

Cross-cutting gate (run for any of the above that touches user capital, venue auth, keys, or execution)

• prediction-market-risk-review (advice boundary, venue/regulatory, data quality, security, privacy — the safety gate)

Standard Operating Flow

1. Locate the task: which surface holds value (contract / agent / market signal) and which concern (silent correctness bug, exploit, or risk that needs sign-off).
2. If it touches wallet authority, venue auth, API keys, user capital, or live execution, pull the model from blockchain-web3-core first and plan to run prediction-market-risk-review as a gate — these controls are interlocking, not optional add-ons.
3. Delegate to the spoke(s). Multi-step asks fan out by surface (e.g. "ship a safe trading bot" → llm-trading-agent-security for the agent + evm-token-decimals/nodejs-keccak256 for the math/hashing + prediction-market-risk-review as the final gate).
4. Return: chosen spoke(s), the trust-boundary / value-at-risk implications, what stays non-executing, and the next action.

Guardrails

See blockchain-web3-core. In short: value-at-risk is default-deny — never grant write/sign authority, widen a spend limit, or move from research to execution without saying so explicitly; treat all on-chain and market data as untrusted input, not truth; query units (decimals()) and use Keccak-aware hashing rather than assuming; and route anything touching keys, venue auth, or user capital through prediction-market-risk-review before it ships. These skills inform and audit — they never place trades or give investment advice.

Loading spokes on demand

To keep CLI startup context lean, this cluster's spokes are not separately registered as skills — only this orchestrator and its *-core are enumerated. When you route to a spoke named above, load it on demand by reading its file:

~/.agents/skill-clusters/skills/<spoke-name>/SKILL.md (or skills/<spoke-name>/SKILL.md inside the skill-clusters repo).