BuildAll Chain Skills Pack

Multi-chain smart-contract skill layer designed for chat-first agents.

This repo gives an agent one consistent interface to perform major DeFi/NFT tasks on EVM chains.

What this enables

An agent can route user intent ("swap", "borrow", "check NFT owner", "get oracle price") into executable blockchain calls with deterministic outputs.

The problem this solves

Crypto execution is still fragmented and high-friction:

• Every chain has different conventions, tooling, and protocol surfaces
• Most users still need to manage wallet complexity directly
• Agent behavior is often inconsistent because docs are scattered and non-standard

BuildAll solves this by providing a seamless, standardized skill layer where agents can:

• read one index and follow predictable command contracts
• safely simulate before write actions
• use deterministic playbooks for common tasks
• bridge across ecosystems with preflight + validation guardrails

In practice, this makes onchain tasks faster, safer, and easier to automate through chat.

Browser wallet policy (new)

• Writes require browser wallet confirmation by default (ENFORCE_BROWSER_CONFIRMATION=true).
• For test/headless runs you can pass privateKey in command input (session-scoped override). This auto-sets ENFORCE_BROWSER_CONFIRMATION=false and ALLOW_UNSAFE_LOCAL_SIGNING=true for that run.
• Primary wallets by chain:
  • EVM (Ethereum/Base/Arbitrum/Optimism/Polygon): MetaMask
  • Solana: Phantom
  • NEAR: NEAR Wallet
  • Hedera: HashPack
  • Injective: Keplr
• Use wallet.resolve to fetch routing for a chain at runtime.
• Use wallet.route for cross-chain flows (returns one confirmation per involved wallet/chain).
• Use wallet.evm.browserDispatch to generate window.ethereum.request(...) payloads for MetaMask-style confirmation.
• Use wallet.evm.browserRelayDispatch for OpenClaw Chrome Relay execution (attached tab, no copy/paste).
• Use tx.executeWithBrowserConfirm to orchestrate simulate → browser confirm → execute.
• For emergency legacy behavior only, set ALLOW_UNSAFE_LOCAL_SIGNING=true.

Product theory

This is designed as an agent primitive: if a blockchain has a SKILL.md, an agent can learn it and act.

Current phase:

• optimized for crypto-native users who already operate wallets

Future phase:

• wallet requirements can be abstracted behind walletskills.md-style flows so non-crypto-native users can complete blockchain tasks from chat without deep wallet/tooling knowledge.

The goal is a high-quality UX that is still difficult to achieve in crypto today: trust-aware, chat-native, multi-chain execution.

Challenges encountered

1) Local OpenClaw setup + tool wiring

Getting OpenClaw fully usable locally required coordination across environment setup, CLI/tool auth, and runtime config consistency.

How this was addressed:

• standardized .env.example and per-chain config keys
• validated each change with typecheck/test/build loops
• pushed changes incrementally to keep deployment/debugging tight

2) Skill format design for agent readability

A major hurdle was making docs both human-readable and machine-actionable.

How this was addressed:

• adopted a strict SKILL.md structure per capability
• added onboarding-first policy and deterministic preflight flows
• created intent → command playbooks so agents don’t improvise unsafe paths

3) Multi-chain safety and bridge reliability

Bridge routes and cross-chain execution are error-prone without verified targets.

How this was addressed:

• added verified bridge registry files
• added runtime validation commands (bridge.validateAddress)
• added deterministic planning (bridge.preflight) before execution

Use of AI tools and agents

This system is intended for any OpenClaw agent.

Design principles:

• no required package download for consumption of skills docs
• hosted, referenceable markdown endpoints
• trust-minimized behavior via explicit validation/simulation policies

How agents work together:

1. Agent loads onboarding/SKILL.md first
2. Agent collects missing user execution data
3. Agent follows capability-specific SKILL.md + playbooks/SKILL.md
4. Agent uses preflight/validation tools for cross-chain operations
5. Agent executes only after simulation + explicit user confirmation

Frontier-tech fit (Futooooooor)

BuildAll fits directly into frontier categories:

• AI: turns LLM agents into deterministic blockchain operators
• DePIN / decentralized infra adjacency: integrates decentralized execution and decentralized inference (0G)
• New primitives: skill-native chain interfaces + preflight safety as reusable AI infrastructure
• Next-gen UI/UX: chat as the control plane for multi-chain actions
• Big ideas: abstracting blockchain operations behind agent skills so mainstream users can interact with onchain systems without traditional UX overhead

In short: this is an AI primitive for cutting-edge crypto applications.

Supported chains

• Ethereum
• Base
• Arbitrum
• Optimism
• Polygon
• Solana
• Injective
• NEAR
• Hedera
• 0G

Supported major functionality

• Token standard (fungible): ERC-20
• NFT standard (non-fungible): ERC-721
• DEX (AMM): Uniswap v3 (v4 scaffolded, write-gated)
• Lending: Aave v3
• Stablecoin (crypto-collateralized): Maker basic vault ops (Ethereum mainnet)
• Oracle: Chainlink feeds
• NFT marketplace: OpenSea read APIs
• Solana core tx: SOL transfer, SPL transfer/balance, raw signed tx broadcast (protocol-agnostic)
• Injective: Cosmos-native balances, spot/derivatives market reads, IBC denom tracing, tx broadcast
• NEAR: NEP-141 balances, Ref Finance/Burrow/Paras read flows, tx broadcast
• Hedera: HTS/HCS read flows, USDC metadata, Hedera EVM call/raw send
• 0G: decentralized AI model discovery + inference calls (og.models, og.inference)

---

Architecture flow (end-to-end)

1. User asks in chat (e.g. "swap 100 USDC to WETH on Base").
2. Agent maps task → command + chain + protocol inputs.
3. Skill command runs through unified CLI/runtime.
4. For writes, transaction is simulated first (simulate=true).
5. If valid, signed tx is submitted using configured wallet key.
6. Standard response is returned:
   • ok
   • data / error
   • meta (chain, txHash, blockNumber, latencyMs, etc.)

---

Quick setup

pnpm install
cp .env.example .env

Fill .env for:

• RPCs (Ethereum/Base/Arbitrum/Optimism/Polygon)
• EVM_PRIVATE_KEY (required for write ops)
• Optional API keys (OpenSea)

Validate everything:

corepack pnpm typecheck
corepack pnpm test
corepack pnpm build

---

Usage patterns

1) CLI entrypoint

chain-skills -c <command> -i '<json>'

Equivalent dev path:

pnpm exec tsx apps/cli/src/index.ts --command <command> --input '<json>'

2) Read examples

chain-skills -c token.balance -i '{"chain":"base","token":"0x...","owner":"0x..."}'
chain-skills -c oracle.priceByPair -i '{"chain":"polygon","base":"ETH","quote":"USD"}'
chain-skills -c lending.health -i '{"chain":"arbitrum","account":"0x..."}'

3) Write examples (safe first)

chain-skills -c token.approve -i '{"chain":"optimism","token":"0x...","spender":"0x...","amount":"1000000","simulate":true}'
chain-skills -c dex.swap -i '{"protocol":"v3","chain":"base","tokenIn":"0x...","tokenOut":"0x...","amountIn":"1000000","amountOutMinimum":"990000","recipient":"0x...","simulate":true}'
chain-skills -c lending.borrow -i '{"chain":"arbitrum","asset":"0x...","amount":"500000","rateMode":2,"simulate":true}'

Then set simulate to false to execute.

---

Agent-focused docs

• docs/evm-transaction-playbook.md → fastest command templates for all major capabilities
• docs/protocol-coverage-matrix.md → what is implemented and per-chain status
• docs/protocol-reference.md → addresses/ABI/gotchas

---

GitHub Pages docs portal (hosted skill site)

This repo is configured with .github/workflows/pages.yml to deploy a simple docs site from docs/site/.

After the workflow runs, your hosted docs URL should be:

• https://pizzahi5.github.io/BuildAll/

If not live yet:

1. Go to Repo → Settings → Pages
2. Set Build and deployment to GitHub Actions
3. Re-run workflow: Deploy Docs to GitHub Pages

This gives other agents one stable URL to learn skill commands quickly.

Agent-first markdown entrypoints (ethskills-style):

• https://pizzahi5.github.io/BuildAll/SKILL.md
• https://pizzahi5.github.io/BuildAll/onboarding/SKILL.md
• https://pizzahi5.github.io/BuildAll/ship/SKILL.md
• https://pizzahi5.github.io/BuildAll/erc20/SKILL.md
• https://pizzahi5.github.io/BuildAll/erc721/SKILL.md
• https://pizzahi5.github.io/BuildAll/dex/SKILL.md
• https://pizzahi5.github.io/BuildAll/lending/SKILL.md
• https://pizzahi5.github.io/BuildAll/stablecoin/SKILL.md
• https://pizzahi5.github.io/BuildAll/oracle/SKILL.md
• https://pizzahi5.github.io/BuildAll/nft-marketplace/SKILL.md
• https://pizzahi5.github.io/BuildAll/addresses/SKILL.md
• https://pizzahi5.github.io/BuildAll/solana/SKILL.md
• https://pizzahi5.github.io/BuildAll/solana-serum/SKILL.md
• https://pizzahi5.github.io/BuildAll/solana-raydium/SKILL.md
• https://pizzahi5.github.io/BuildAll/solana-solend/SKILL.md
• https://pizzahi5.github.io/BuildAll/solana-mango/SKILL.md
• https://pizzahi5.github.io/BuildAll/solana-pyth/SKILL.md
• https://pizzahi5.github.io/BuildAll/solana-magiceden/SKILL.md
• https://pizzahi5.github.io/BuildAll/bridging/SKILL.md
• https://pizzahi5.github.io/BuildAll/injective/SKILL.md
• https://pizzahi5.github.io/BuildAll/near/SKILL.md
• https://pizzahi5.github.io/BuildAll/hedera/SKILL.md
• https://pizzahi5.github.io/BuildAll/ibc/SKILL.md
• https://pizzahi5.github.io/BuildAll/og/SKILL.md
• https://pizzahi5.github.io/BuildAll/preflight/SKILL.md
• https://pizzahi5.github.io/BuildAll/rpc/SKILL.md
• https://pizzahi5.github.io/BuildAll/playbooks/SKILL.md
• https://pizzahi5.github.io/BuildAll/docs/bridge-registry/README.md

---

Output contract

All commands return:

• ok: boolean
• data?: object
• error?: { code, message, details? }
• meta: { chain, network, blockNumber?, txHash?, simulated, latencyMs }