skills/cli-cast/SKILL.md
Use for Foundry cast/on-chain actions: prepare, simulate, sign, or broadcast transactions; call contracts; sign messages; encode/decode ABI/calldata; or make RPC calls. For read-only balances/history, prefer evm-atlas.
npx skillsauth add paulrberg/agent-skills cli-castInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Separate read, preparation, simulation, signing, and broadcast so no state-changing action is hidden inside command construction.
Resolve the target chain through evm-atlas. Use RouteMesh only when that chain is supported and ROUTEMESH_API_KEY is
available:
https://lb.routeme.sh/rpc/CHAIN_ID/ROUTEMESH_API_KEY
Otherwise use the target chain's verified public RPC. Read references/chains.md only as a
limited fallback when evm-atlas is unavailable. Never infer Ethereum when the chain is ambiguous.
Read-only calls, transaction/receipt inspection, ABI encoding/decoding, selector lookup, ENS resolution, and RPC queries
may run without a transaction approval. Prefer evm-atlas for balances, holdings, transfers, and history.
Use current cast <command> --help for exact syntax. Typical operations:
cast call "$CONTRACT" 'balanceOf(address)(uint256)' "$ACCOUNT" --rpc-url "$RPC_URL"
cast tx "$TX_HASH" --rpc-url "$RPC_URL"
cast receipt "$TX_HASH" --rpc-url "$RPC_URL"
cast calldata 'transfer(address,uint256)' "$TO" "$AMOUNT"
cast decode-calldata 'transfer(address,uint256)' "$CALLDATA"
For two or more compatible reads on one chain, batch through Multicall3 at 0xcA11bde05977b3631167028862bE2a173976CA11.
Do not batch calls whose result depends on the original msg.sender.
Resolve and validate chain ID, sender, target, function signature, arguments, calldata, native value, nonce, and fee assumptions without signing. Do not request or load key material during preparation.
Simulate the exact prepared call from the intended sender and value, then estimate gas. Use a fork or project simulation
when available; otherwise use read-only cast call, cast estimate, and trace facilities. A successful simulation is
evidence, not authorization to sign.
Before any signature or broadcast, present one concrete review containing:
Stop and require explicit user confirmation of this review in a subsequent message. If any reviewed field changes, simulate again and present a revised review.
Read references/browser-signing.md only after approval when a browser wallet is available. Prefer browser, encrypted keystore, or hardware wallet in that order. Use an environment-backed private key only when the user explicitly opts in or no safer method is available; never ask for a key in chat or print it.
cast send signs and broadcasts in one command. Run it only after the review approval. Signing a message or typed data
also requires a review of the exact payload, domain, chain binding, and intended use before approval.
After broadcast, capture the transaction hash and verify the receipt on the reviewed chain. Report status, block, gas used, and the explorer link. Do not retry a failed or uncertain broadcast without first checking whether the transaction exists.
Stop before signing when the signer, sender, chain, target, decoded intent, cost ceiling, or simulation result is unresolved. Stop before retrying when broadcast outcome is ambiguous. Completion requires either a verified read result, an approved signature artifact, or a mined receipt matching the reviewed transaction.
development
Use when creating or substantially redesigning web interfaces, landing pages, dashboards, components, or other frontend UI where visual direction and implementation quality matter. Produces subject-specific art direction, accessible responsive code, and rendered visual verification.
development
Orchestrate one to five Sonnet subagents to implement an approved Claude Code plan.
tools
Open the CoinGecko historical-data page for a coin/date in Chromium via Chrome DevTools MCP.
tools
Orchestrate one to five Codex CLI agents to implement an approved Claude Code plan.