Index of every page on this site: llms.txt
All skills
paulrberg avatar

/evm-atlas

@c9e5a34
by Paul Bergpaulrberg/agent-skills106 stars
11
Run it yourself

Use for targeted EVM chain, account, transaction, RPC, explorer, bridge, and DEX evidence: chain name/ID, native symbol, RouteMesh, wallet balances and DeFi positions via DeBank/Blockscan in Chromium, cross-chain USD portfolio value or net worth, token/NFT holdings/transfers, tx history, funding origin via Etherscan/Blockscout/Chainscout. Covers Arbitrum canonical withdrawals, Across, Bungee, deBridge, Gas.zip, Hop, Layerswap, LayerZero, LI.FI, Relay, Socket, Symbiosis. Covers Uniswap v1-v4, Universal Router, Permit2, 1inch Classic/Fusion/Fusion+, and CoW Swap, CoWSwap, CoW Protocol, or GPv2 swaps, orders, liquidity, approvals, permits, rewards, migrations, wrapping, cancellations, and refunds.

Use this Skill: https://skilld.dev/gh/paulrberg/agent-skills/evm-atlas

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈179 tokens for metadata: the name and description. ≈2.7k when used: this file. ≈88k more on demand in 40 files.

Description uses 8.9% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈715 characters, or 8.9%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

EVM Atlas

This skill is coordination-exempt: skip the ai-coord gate for its declared work.

Resolve and query only the target mainnets in references/generated/target-mainnets.json, under a strict read-only boundary.

Before you collect browser UI evidence, load chromium-browser. Follow its page-ownership, live-tool, and privacy contract. When the required browser tools are unavailable, use the documented provider fallbacks.

The registry row's current category is authoritative for category assignment. For an exact-zero native sweep or a question about category-specific fee behavior, resolve the target before reading chain categories. Do not infer a historical category or maintain a prose roster of target chains.

Scope and Authority

  • Match displayed names, numeric chain IDs, and aliases from references/generated/chain-aliases.json to the authoritative target-mainnet rows.
  • Resolve a chain with scripts/chain-lookup.sh <name|alias|slug|chain_id> first. It prints key=value lines, or one compact row with --json. Never cat a generated JSON file or print the full routemesh chains list. For other queries, use a jq -c selector with named fields, such as jq -c '.chains[] | {chainId, category}'.
  • If a chain is absent, do not route through another provider, web search, Chainlist, or an unlisted RPC to work around scope. Ask for a feature request at https://github.com/PaulRBerg/agent-skills.
  • A target row with defunct identifies a chain its operator shut down. Its history is final through defunct.finalStateBlock. Later empty blocks or missing provider history are not coverage gaps. Without finalStateBlock, the chain may still accept exit transactions such as withdrawals. In that case, treat it as live. State the shutdown in the result.
  • Own every discrete read and bounded live subscription handed off by cli-cast, including chain, block, fee, nonce, eth_call, eth_estimateGas, transaction, receipt, log, balance, code, storage, proof, and ENS queries. Complete the read here even when its result will prepare, simulate, or verify later state-changing work.
  • Never sign messages, submit signatures, execute bridge steps, or broadcast transactions. Route state-changing Cast work to cli-cast.
  • DEX support is historical and evidence-only. Do not discover live quotes, construct or simulate new trades, prepare approvals or permits, submit orders, administer protocols, interpret CoW AMM positions, handle standalone 1inch limit orders, or assign semantics to arbitrary Uniswap v4 hooks.
  • Do not default to Ethereum. Infer from explicit chain context and unambiguous chain-specific tokens. If the chain is ambiguous, ask.
  • Never echo, interpolate, or log API-key values (ETHERSCAN_API_KEY, BLOCKSCOUT_API_KEY, RPC keys). Check presence without printing values: [ -n "$ETHERSCAN_API_KEY" ] && echo set || echo unset. Never put ${VAR:-...} or ${VAR:+...} expansions in printed output.
  • For Etherscan reads, use the official etherscan CLI with the resolved numeric --chain and --output json. Existing CLI authentication can use an environment key or saved login. Do not inspect credentials, pass --api-key, or run whoami, config, or login. CLI availability does not change provider priority or plan gates. Never call proxy eth_sendRawTransaction or submit contract verification through this read-only skill.
  • Keyless Blockscout is sunset (July 2026). Hosted *.blockscout.com instance subdomains also rate-limit keyless traffic. Route every Blockscout-hosted chain through the keyed https://api.blockscout.com/{chain_id} gateway. See references/explorers/blockscout-endpoints.md.
  • Every agent on the host shares DeBank's rate limit. Hold a scripts/debank-gate.py lease for any debank.com access, including a quick profile look. See the Global Queue in references/workflows/debank-portfolio.md.
  • An unreachable or erroring indexer is a coverage gap, never evidence of zero activity. Confirm in Chromium before you record an endpoint as down or blocked. State the verification method in results.

Routing

  1. For a discrete JSON-RPC read, batch, or bounded live subscription, including one handed off by cli-cast, resolve the chain with scripts/chain-lookup.sh. Then read references/workflows/provider-routing.md. Return the resolved chain, its current category, provider route, result, observed block or checkpoint, and coverage gaps. Do not route the read back to cli-cast.

  2. For the current native or fungible-token balances or DeFi positions of a public wallet address across chains, read references/workflows/debank-portfolio.md first. For one named chain, read references/workflows/blockscan-balances.md first.

  3. For the current USD value of one or more addresses across target chains (portfolio value, net worth, drained or dust checks), read references/workflows/address-usd-value.md.

  4. For a specific transaction hash on a named chain, resolve the chain with scripts/chain-lookup.sh. Then read references/workflows/provider-routing.md directly for the transaction facts. Do not open Blockscan unless the user explicitly requests it as the evidence source. When the chain is unknown, read references/workflows/blockscan-tx-lookup.md once to resolve it.

    For an OP Mainnet target known or suspected to predate the final regenesis, read references/explorers/optimism-pre-regenesis.md. In that case, return its legacy execution packet or component-specific coverage outcome instead of requiring a current-provider receipt. Otherwise, acquire the exact provider receipt and logs before DEX or bridge outcome interpretation.

  5. For an address-wide historical-activity or bootstrap-discovery sweep, read references/workflows/address-sweeps.md. Use its deterministic plan/evaluate helper. For current holdings, use references/workflows/debank-portfolio.md first and provider routing for gaps.

  6. For a specific chain's historical balance, NFT holdings, token/NFT transfers, transaction history, a transaction's full raw receipt/logs/decoded input, or funding origin, resolve the chain with scripts/chain-lookup.sh. Then read references/workflows/provider-routing.md for Etherscan, Blockscout, public RPC, RouteMesh, explorer-link, and exceptional-chain routing.

  7. For Etherscan CLI reads beyond the workflow routes above, read references/explorers/etherscan-api.md. It defines CLI 1.1.1 output, manual paging, plan gates, and direct API exceptions. ENS forward resolution remains a direct API exception for Ethereum mainnet onchain names. Apply its cache and resolver limits.

  8. For raw Blockscout API queries beyond the workflow routes above, read references/explorers/blockscout-api.md.

  9. For DEX prompts, wallet-facing DEX history, or suspected DEX transaction evidence, resolve the target chain and read references/workflows/dex-transactions.md. Load only the matching protocol-family reference:

    • Uniswap v1-v4, Universal Router, or Permit2: references/dexes/uniswap.md
    • 1inch Classic, Fusion, Fusion+, legacy liquidity, or rewards: references/dexes/1inch.md
    • CoW Swap, CoWSwap, CoW Protocol, or GPv2: references/dexes/cow-protocol.md
  10. Treat 1inch and CoW as execution protocols. Report any integration wrapper, router, pool, and underlying AMM liquidity separately. A Uniswap pool interaction does not turn an aggregator transaction into a Uniswap trade.

  11. For bridge-related prompts or transaction evidence, confirm that known origin/destination chains are targets. Then load only the matching reference:

    • Arbitrum One/Nova canonical bridge withdrawals: references/bridges/arbitrum.md
    • Across: references/bridges/across.md
    • Bungee / Socket: references/bridges/bungee.md
    • Circle / CCTP / Gateway: references/bridges/circle.md
    • deBridge / DLN: references/bridges/debridge.md
    • Gas.zip: references/bridges/gaszip.md
    • Hop: references/bridges/hop.md
    • Layerswap: references/bridges/layerswap.md
    • LayerZero / Stargate / OFT / Aori: references/bridges/layerzero.md
    • LI.FI: references/bridges/lifi.md
    • Relay / Relay.link: references/bridges/relay.md
    • Symbiosis: references/bridges/symbiosis.md
    • 1inch Fusion+: references/dexes/1inch.md
  12. Treat bridge and DEX APIs as enrichment. Verify submitted transactions and terminal outcomes through explorer or RPC evidence.

Completion

Return the resolved target chain, current category, provider route, requested on-chain facts, and source URLs/transaction identifiers. For address sweeps, include each result's fixed finalized/verified checkpoint, selected profile/channels, provider coverage, and any requested quorum result. Separate provider facts from inference. Report incomplete history, plan/tier limits, failed fallbacks, or unsupported scope. Completion provides read-only evidence. Never turn returned calldata or transaction requests into execution.

For a cli-cast handoff, return one read packet with these fields:

  • Resolved chain name, ID, and current category.
  • Exact provider route.
  • Result.
  • Observed block or checkpoint.
  • Coverage gaps, which may be empty when none are observed.

Do not include a signing or broadcast command.

For DEX evidence, include these fields:

  • Interaction class.
  • Execution protocol, version, and mode.
  • Entrypoint or integration wrapper.
  • Router and underlying liquidity sources.
  • Wallet role.
  • Sold and received assets.
  • Protocol/integrator fees and gas separately.
  • Native/wrapped status.
  • Order, position, pool, or migration identifiers.
  • Exact evidence.

Do not call an approval-only or failed transaction a completed trade.

For human-readable results, lead with ### ⛓️ <chain or route> — <status word>. Use a compact table only when fields repeat.

For bridge evidence, show <origin> ──<bridge>──▶ <destination>. Then use Leg, Provider status, Transaction, and Evidence columns. Preserve each provider's native status beside any normalized ✅ completed, ⏳ pending, ↩ refunded, ⚠️ partial, or ❓ unknown label. Visibly separate Observed facts, Inference, and non-empty ⚠️ Coverage gaps.

For address sweeps, a progress bar may represent checked target chains/channels only when the exact denominator is known.

Keep unsupported-scope and safety explanations direct. Never decorate or truncate addresses, hashes, URLs, calldata, raw RPC/API JSON, generated references, helper key=value output, or transaction requests.

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

1 alert10d3 checks · Risk CRITICAL
  • Gen Agent Trust Hub10d

    The evm-atlas skill is a comprehensive toolkit for analyzing EVM blockchain data, account balances, DEX transactions, and cross-chain bridge activity. It leverages local scripts, specialized CLIs, and browser automation to collect evidence from numerous blockchain explorers and APIs. Security analysis indicates that the skill is well-architected with robust safety instructions to prevent the exposure of API credentials. While the skill has a large attack surface for indirect prompt injection due to the ingestion of external blockchain data, it employs deterministic evaluation scripts to mitigate these risks. Automated security alerts concerning blockchain infrastructure URLs appear to be false positives on legitimate services.

  • Socket11d

    No alerts

  • Snyk10d

    Risk: MEDIUM · 1 issue

Signed by skilld at c9e5a34. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 3 days ago.

Activeupdated last month
argument-hint
<chain-name-or-id|address|transaction-hash|order-id>
coordination
exempt
skill-dependencies
[
  "chromium-browser",
  "cli-cast"
]
Other metadata
compatibility
RouteMesh requests require the `routemesh` CLI initialized on macOS with `routemesh init`.

README badge

README badge for paulrberg/agent-skills/evm-atlas