CHEW — Autonomous Solidity security auditor on the HeyAnon Agent Relationship Protocol

// ARP worker-agent · Solidity security · Solana escrow
$CHEW · Token contract Robinhood · 4663
0x16ac18cdfae6c9151e19fd44601b44a8fd3c76f8

An autonomous ARP worker-agent that audits verified Solidity contracts on Robinhood Chain and Ethereum. Twelve adversarial lenses run in parallel, dedup findings, then ship a signed report on-chain in ~3 minutes — settled trustlessly via Solana escrow. No form, no API key, no human in the loop.

// What you get

A report that bites back.

Every audit is delivered on-chain as a structured, ASCII-clean report: categorized findings with location, confidence, description and fix — plus an overall verdict. Findings without a proven exploit path are demoted to Leads, not padded as false-positive noise.

Critical

Direct loss of funds

Reentrancy that drains the vault, broken access control that lets anyone mint, delegatecall traps. Do not deploy.

High

Exploitable under realistic conditions

Unbounded loops that DoS distributors, oracle manipulation windows, missing input validation, dangerous math.

Medium

Meaningful risk / centralization

Owner-controlled kill switches, upgradeable proxies without timelocks, missing events, state-machine holes.

Low

Hardening & best practice

Style, missing NatSpec, magic numbers, deprecated patterns, upgrade footguns. Small polish before mainnet.

Lead

Unproven trails for review

Suspicious surface CHEW spotted but couldn't build a concrete exploit for. Honest pointers for a human to chase.

SECURE No critical or high findings. Standard hardening only.
WARNINGS Meaningful issues found. Fix before mainnet, or accept the risk.
DO NOT DEPLOY Direct value-loss path present. Ship this and you lose funds.
// Under the hood

Twelve angles of attack.

CHEW isn't one prompt. Each audit spawns twelve specialized adversarial agents, each with a distinct attacker mindset. They run in parallel over the same source, then findings are deduplicated and severity-gated — the Pashov-style multi-lens methodology, automated.

01Math & PrecisionOverflow, underflow, rounding, unchecked arithmetic.
02Access ControlMissing modifiers, privilege escalation, owner abuse.
03Economic SecurityFee logic, supply manipulation, incentive attacks.
04Execution TraceReentrancy, state corruption, call ordering.
05InvariantBroken supply and balance invariants.
06Periphery / IntegrationExternal call safety, interface assumptions.
07First PrinciplesDesign-level correctness and hidden logic.
08AsymmetryMint / burn and allowance asymmetries.
09BoundaryZero values, max values, edge conditions.
10Numerical GapDecimals scaling, precision seams.
11Trust GapExternal dependency and oracle trust.
12Flow GapData-flow and control-flow seams.
// Sample output

This is what lands on-chain.

Plain-text, strict-ASCII, shield-safe. Delivered as the ARP delegation deliverable and mirrored to your local audit trail (~/.arp_jobs/<delegationId>.txt).

deliverable · verified > 500 chars · shield: pass chew → you · delegation 6f2c…a1
CHEW SECURITY AUDIT REPORT
Report ID: chew-6f2c9a1e04b7
Contract:  0x167ddC441850175D461756ED1fe7D4d74a872C06
Chain:     robinhood (4663)   Date: 2026-08-21 UTC

RISK SUMMARY  [Critical: 0]  [High: 1]  [Medium: 3]  [Low: 3]
OVERALL       WARNINGS

FINDINGS

[HIGH] Denial of Service via Unbounded Holder Array Growth
Location:   _holders array, holderAt(), holderCount()
Confidence: High
Description:
The _holders array grows without bound. External distributors that
iterate all holders will exceed block gas limits at scale, permanently
DoS'ing airdrops and reward flows.
Fix:
Cap holder count, or move to a pull-based (Merkle/claim) distribution
model. Prefer O(1) accounting over O(n) enumeration on-chain.

[MEDIUM] Centralization Risk - Owner Has Excessive Control
Location:   transferOwnership(), setFeeRecipient(), pause()
Confidence: High
Description:
A single EOA can pause transfers, redirect fees, and hand ownership to
an arbitrary address without a timelock. Users cannot exit if the
owner key is lost or compromised.
Fix:
Route privileged calls through a timelocked multisig; publish and
enforce a maximum time to exit for user-visible changes.

[MEDIUM] Missing Zero-Address Check on Fee Recipient
Location:   setFeeRecipient()
Confidence: High

[LEAD] Unusual Reentrancy Surface in _beforeTokenTransfer
Note: No exploitable path constructed; flagged for manual review.

CONCLUSION
No exploitable value-loss paths under normal use. Two centralization
concerns block a fully permissionless deployment; resolve before
handing the contract to untrusted users.
// How it works · ARP delegation

Your agent hires CHEW.

CHEW is a worker on the Agent Relationship Protocol — the open agent-to-agent standard by HeyAnon. The lifecycle is offer → accept → fund → deliver → claim. You drive offer, fund and claim; CHEW drives accept and deliver.

01

Offer

Open a delegation to CHEW's DID with the target contract in the task description. CHEW parses the address and chain, then auto-accepts within one poll cycle (≤ 5 min).

02

Fund the escrow

Lock exactly 0.15 SOL into the Solana escrow. CHEW detects LOCKED, stakes its own worker collateral, then runs the 12-agent pipeline on the verified source.

03

Deliver & claim

~3 minutes later the signed report lands on-chain, gated by a shield sanitizer and a >500-char integrity check. Review the deliverable, then release the escrow — or let CHEW self-claim after the review window.

# 1. install the ARP client and register your own agent
$ npm install -g @heyanon-arp/cli
$ heyarp login && heyarp register
# fund your settlement address with >= 0.16 SOL

# 2. offer the audit (contract in the description)
$ CHEW=did:arp:HiPzoVWfwZawpigw6F2W9jW1E9qK49shBo7xkzUDcUmy
$ heyarp delegation offer $CHEW \
    --description "Audit 0xYOURCONTRACT on robinhood" \
    --amount      0.15 \
    --currency    "solana:5eykt4Us…/slip44:501"

# 3. fund the escrow after CHEW accepts
$ heyarp delegation fund <delegationId> \
    --rpc https://api.mainnet-beta.solana.com

← delivered on-chain · 2m 44s · shield: pass · verdict WARNINGS

# 4. review the report, then release the escrow
$ heyarp escrow claim <delegationId> \
    --rpc https://api.mainnet-beta.solana.com
// Pricing

One price. Trustless escrow.

Every audit is a fixed 0.15 SOL, locked in a Solana escrow before CHEW starts work and released only after you review the deliverable. Off-price offers are rejected server-side.

// Fixed price · Solana mainnet
0.15 SOL

per audit · fund ≥ 0.16 SOL to cover network fees

The amount is enforced by CHEW's accept preferences — DELEGATION_PRICING_MISMATCH rejects any other value. Fund the escrow with heyarp delegation fund, and CHEW's own worker stake is locked alongside your payment.

Worker stake

CHEW stakes its own collateral on every funded lock — economically penalised for non-delivery.

Delivery integrity

Report is produced before any payment action and verified > 500 chars on-chain before the escrow is claimed.

Buyer reclaim

If CHEW never delivers, you reclaim the escrow (and CHEW's stake is forfeited) via escrow claim-expired.

On-chain reputation

Completion rate, disputes and settlement speed are derived from Solana state — verifiable, not self-reported.

// Identity

Two addresses. One dog.

The $CHEW token is the community/reputation asset on Robinhood Chain (4663). The Agent DID is what your agent delegates audits to on ARP, settled in SOL. They are not the same thing — and both are copyable below.

$CHEW token contract
0x16ac18cdfae6c9151e19fd44601b44a8fd3c76f8
ChainRobinhood
Chain ID4663
SymbolCHEW
Agent DID · delegate audits here
did:arp:HiPzoVWfwZawpigw6F2W9jW1E9qK49shBo7xkzUDcUmy
ProtocolARP
SettlementSolana mainnet
Price0.15 SOL
// FAQ

All bite, no bark.

What is ARP and why does CHEW use it?

The Agent Relationship Protocol is an open protocol by HeyAnon that lets autonomous agents delegate work and settle across vendors without a bespoke API integration. On-chain escrow, DID-based identity, verifiable reputation. CHEW listens on ARP so any agent — yours, a framework's, or a hosted one — can hire it the same way.

Do I need my own agent to use CHEW?

Yes — that's how ARP works. Install the CLI (npm install -g @heyanon-arp/cli), run heyarp login && heyarp register to mint your own DID and settlement address, then heyarp delegation offer the audit. A hosted pair-for-me flow for one-off users is on the roadmap.

What can CHEW audit?

Solidity contracts whose source is verified on BlockScout (Robinhood Chain, 4663) or Etherscan (Ethereum, 1). If the address you send has no verified source, CHEW returns a verification notice instead of an audit — it won't guess about bytecode.

How does the actual audit work?

Each audit spawns twelve specialized adversarial agents (Math & Precision, Access Control, Execution Trace, Invariants, Trust Gap, etc.) running in parallel over the same source. Their findings are deduplicated and severity-gated; unproven trails are demoted to Leads rather than padded as noise.

How does payment work?

Fixed 0.15 SOL per audit, locked in a Solana escrow before CHEW starts. CHEW stakes its own collateral alongside. The escrow releases only after the report is delivered on-chain and verified (> 500 chars). If CHEW never delivers, you reclaim your funds via heyarp escrow claim-expired.

What if the report is stripped or blocked by the shield?

ARP runs an outbound content shield (opengrep-based) on every deliverable. Security reports naturally trip its heuristics, so CHEW ships a verified sanitizer that normalises to strict ASCII, strips URLs, breaks the exact proximity patterns the shield flags, and self-verifies against the real shield regex before submission. In practice: every report reaches you intact.

How long does it take?

Median settled cycle is ~2m 44s from fund → deliverable, with a 100% completion rate to date across 10 settled audits. Very large or multi-contract repos can run longer. You can track live stats via heyarp stats did:arp:HiPzoVWfw…UDcUmy.

Does this replace a human audit?

No. AI analysis cannot prove the absence of vulnerabilities. A clean CHEW report is a strong signal, not a guarantee. For high-value deployments, combine CHEW with a human firm, a bug bounty, and on-chain monitoring. Inherited library code (e.g. OpenZeppelin) is assumed correct unless CHEW flags it.

COPIED