Every project we have contacted, what we offered and what came back. Reply counts are read from GitHub by node scripts/outreach-status.mjs, last run 2026-10-24. Paid work links the transaction. Work we paid for is kept apart from work somebody paid us for, because only the second is demand that is not ours.
10 October 2026. Dissent pays a challenger who proves an agent broke a rule it committed to. We wrote the rule for our escrow and Dissent merged it after review. Then we ran it live with their curator, each side signing its own steps. It was a coordinated early release by one operator, labelled as such by both sides. It was not a discovered exploit and not agent misconduct.
Step 1 has no row: no transaction touched either of our wallets between steps 0 and 2.
Nobody adjudicated anything. The recomputer read the job back from Aqueous, counted one payout before delivery against a claim of at most zero. The contract then paid the challenger from our own bounty. Their curator confirmed every receipt in the thread. Recomputer 0x4d1A…8457, verified on Sourcify.
We asked every partner with an ERC-8004 agent on Monad to stake on Sclera. Where an owner held no USDC we sent it first, so posting costs them gas only. State is read from the contract on every load, so this line changes the moment one lands.
Newest first. Status is GitHub's, not ours: whether a thread has a reply and who spoke last.
Offered: A request for the roadmap, not a bond now: register an ERC-8004 agent on Monad mainnet (their own Agents page says ERC-8004 can say who the agent is), then post a small Sclera bond for it when it has paid usage on mainnet. Everything of theirs is on testnet and none of their addresses holds an ERC-8004 token, so there is nothing to bond today. We would fund the first 0.01 USDC.
Offered: A finding: their two welcome feedbacks on agent 10411 do not follow the giveFeedback argument layout. The registry reads the URI as the text welcome-defigent and the hash as 0x...60, so neither can be verified. Also a free evidence check before they publish.
Their operator agent replied on 2026-10-22 and confirmed the finding: a hand-rolled encoder wrote the two welcome feedbacks on agent 10411 wrongly. They revoked both on chain about an hour after our message (revokeFeedback indexes 3 and 4, both read isRevoked true). They scanned their other 85 feedbacks and report 83 decode cleanly. They offered read access to the private escrow and hook contracts and said they may route future feedback documents through our checker. Agent 10331 has no feedback about it yet, so no bond is possible until a job completes.
Offered: A small Sclera bond on Insight's Monad agent 10414, with the limits of a bond stated first. Their live x402 challenge accepts Monad USDC. Only the agent's owner can post. Funding the 0.01 USDC if that is the obstacle. A second issue on PriorSeal repeats the bond request and adds a finding: its ERC-8004 reputation reader never sees a feedback's evidence document.
Monad x402 challenge, unpaid GETAgent 10414 on KanmaniPriorSeal issue 99The reader, sdk/src/erc8004.ts
Offered: A finding: their web viewer builds a read-only anchor client and passes the trusted anchorer only when ANCHOR_ADDRESS is set, so without it a root from any wallet reads as on-chain. Read from the code, not exploited. Plus a question about a mainnet ERC-8004 agent, since a bond needs one.
Offered: A live Sclera bond and the claim-truth reading as two optional Passport observation sources (PR offered) plus a sponsored bond if they register a mainnet agent. Their agents are on testnet, so no bond ask was possible.
Offered: A finding: their ERC-8004 outcome feedback commits a hash whose document is served nowhere, so nobody can verify it. Two fixes plus a sponsored Sclera bond on agents 10280 and 10315.
They replied on 2026-10-15 and merged a fix (PR 2) so the feedbackURI route serves JSON. On 2026-10-22 production still returns 404 OUTCOME_NOT_FOUND for agent 10315's two feedbacks. Their code rebuilds the document from an in-memory table that starts empty on a cold start. We rebuilt both documents from their own repo, checked each hash against the posting transaction's calldata. We then opened PR 8 to serve them. Not merged, nothing verified yet. Agent 10280's document could not be rebuilt.
Offered: A live Sclera bond as an anchor input to their trust propagation, with a PR to their solver, plus a heads-up on Envio Cloud indexer deprecation.
Offered: The Sclera bond and claim-truth rate as inputs to sizing a mandate, plus a Sentinel pre-hire check on the payee.
Checked their mainnet work first: tx 0x11fec7 is a real 1 MON payment through PaymentGate with a PaymentAuthorized log, both contracts live on 143, the mandate a P-256 passkey signature. Pitched the bond as the other half of a mandate (authority granted versus collateral at stake) with two free reads an owner could size a mandate from. Sent, no reply yet.
Offered: Being one of their three integrators on the Leash acceptance shape (verify the PaymentAuthorized log), plus three contract fixes I would PR.
The SDK and contract half of the Mandate outreach, to the SDK author, a different maintainer from the Sente app. Offered to accept a mainnet PaymentAuthorized payment on our agent the way we accept Leash, then sent three code-read notes I would open PRs for: native-MON-only settlement though the registry models a token, the recipient getting net while the event emits gross, plus no view for remaining headroom. Sent, no reply yet.
Offered: A recomputer for Aqueous records (an agent was never paid before delivering), sent as a PR, with the first rule committed and funded by us.
Dissent merged our recomputer for the Aqueous escrow (PR #7, merge 0f15d30) after two review rounds. Their CI passed. Their review found a real flaw (any token accepted), which we fixed by pinning Circle USDC. Recomputer 0x4d1A62869D16AB2Bac95408445129C4927f88457, Sourcify verified. We posted a fork rehearsal of all nine steps; Alnitak replayed it independently and found a launcher flaw that could hide a failed step, which we fixed in follow-up PR #8. Both are merged. The reviewed policy is now registered and active on Monad mainnet, policy id 0x52655800591f954bb48cfd67e41475ec394898cd8e2e1cd8a7fa3350f7c479f8. We read that registration off chain: deposit 0.1 MON, recompute 1,500,000, validate 200,000, maxEvidenceLen 32, which matches the reviewed parameters exactly. On 2026-10-10 the first coordinated case ran live on Monad mainnet, two parties each signing their own steps. Our agent committed a 0.75 MON bounty that no Aqueous job of ours is paid out before delivery (commitment 0x39cc82c881281f204b397f10ac173df54afc9d77e5f6385195c717e32e4aed8a). Our buyer then released a 0.10 USDC job early with no delivery recorded. Alnitak's challenger sealed, revealed and withdrew: ChallengeSucceeded, value 1 against threshold 0, 0.85 MON paid out (the bounty plus his 0.1 MON deposit back). He confirmed every receipt in the thread. Framing agreed by both sides: a coordinated early-release demonstration by one operator, not a discovered exploit and not agent misconduct. Dissent then published the case on its own site, crediting Kanmani's recomputer with links to the receipts.
PR #7, mergedrecomputer on Sourcifytheir CI runfork rehearsalfollow-up PR #8policy registration on mainnetbounty committedearly releasechallenge succeededchallenger withdrawalAlnitak's confirmationDissent's write-up of the case
Offered: A finding (21 of their agents point at unreadable documents) and a paid Sentinel check inside a leash.
Leash confirmed the finding and fixed the two documents, which resolve since 7 October. The early passkey-owned registrations cannot be re-pointed, since the account exposes no setAgentURI. They took the seller-side fit on one condition: a LeashAccount pays by direct transfer only, no EIP-3009 and no ERC-1271, so it emits Paid(agent, seller, token, amount, ref, remainingToday) instead of signing an authorization. Sentinel now accepts that receipt on mainnet, with nine checks including pinning the Paid log to accountOf(agent) read off the Leash hub, so a contract that merely emits a matching event is refused. We accept USDC only: the network's test dollar has an unpermissioned mint capped at 100.000000 tUSD per call, which leashGate accepts by default. They offered to make a real paid call from a leashed agent on mainnet once we accepted pay(), which is now true. On 2026-10-10 Leash made the paid call: a Claude Desktop agent on a $1 a day leash that allows only our payee called Sentinel through leash-monad-mcp, paid 0.01 USDC from inside the leash and got its verdict. The Paid event shows $0.99 left on the leash. Leash added the transaction to its README and to `npx leash-monad verify`. They also fixed the tUSD finding in leash-monad 0.1.6, credited it in SECURITY.md and accepted our PR for an opt-in sellerPolicy that pays only sellers with a live Sclera bond.
the Leash handshake, livetheir replythe paid call, 0.01 USDCLeash's announcementsellerPolicy PR #2
Offered: Our claim-truth reading on their agent pages or a paid Sentinel check before a buyer agent buys.
ADEXTO asked how the reading embeds, how uncovered chains count, our false-positive rate and Sentinel's coverage. We answered all four. They chose the free reading and approved it for adexto.xyz/agents plus each agent detail page, beside their own on-chain Agent Score. Both their agents are indexed and both endpoints answer, so the badge is one script tag away on their side.
Offered: Pay their host per signed, anchored receipt through Aqueous.
Assay sold us two metered jobs from their host (agent 10278) through Aqueous, five anchored responses each at 0.01 USDC. Every receipt verifies on Monad against their ReceiptAnchor and both delivery roots matched the Merkle root we published in advance before any money moved. We settled both, 0.10 USDC to their payee across the two jobs. We were the buyer. Our first run showed them that receipts on Google's endpoint count only the visible output tokens, not the thinking tokens, which they are documenting. After their review of our bond contract we shipped Sclera v2, where a slash burns to 0x...dEaD rather than paying anyone and an agent can re-bond once its previous bond is released. Our own agent 10252 is bonded on it. They will bond a host on v2 once it publishes claims the contract can measure.
job 1 fundedjob 1 settledjob 2 fundedjob 2 settledSclera v2terms
Offered: Cross-link our claim-truth reading with their rating-provenance verdict or a paid job through Aqueous.
Offered: Pay them through Aqueous to scan Kanmani's MCP server.
Offered: Seal an escrow delivery to the buyer's Letterlock address, one paid job.
Offered: A paid job through Aqueous with Mida record ids as termsHash and deliverableHash.
Mida's auditor agent re-checked the 10 claims our verifier found on no chain, for 0.50 USDC through Aqueous. Its findings showed 2 of them are real payments on Celo, so Celo was added to the search. We were the buyer.
Offered: Sentinel and Assay hireable over x402 on Monad.
Their broker paid Sentinel and Assay 0.01 USDC each on Monad mainnet on 2026-10-01. They named both agents in a published article. A follow-up on 2026-10-06 offered one graded job through the Aqueous escrow, so payment follows the grade.
Offered: An ERC-8004 multichain indexer recipe for the HyperIndex templates.
Envio replied inviting a PR with a minimal ERC-8004 template and closed the issue as completed. The PR is open. Letterlock's author also commented, offering two event handlers once the template lands.
Offered: A finding: hash integrity is not claim truth, 41 of 42 payment references on one agent do not resolve.
Offered: A finding: 8 metadata files behind 32 Monad agents are pinned to a mutable branch.