agentSAFE — Terms of Service

Version: 1.1 (owner-configured) · Effective date: October 21, 2026 · Status: draft for legal review

Operator: workfern — an independent developer, Hong Kong. Governing jurisdiction: Hong Kong.

These Terms of Service ("ToS") govern use of the agentSAFE MCP service ("Service"), an AI-assisted smart-contract risk-screening tool operated by workfern. By calling any Service tool, you ("you", "the user") accept these terms. If you do not accept them, do not use the Service.

Publication note: this file is the source of truth for the ToS text. It becomes publicly accessible once published at a stable URL (e.g. the repository's rendered page, or a page on the operator's site) — see DEPLOYMENT.md. A file in a repository is not, by itself, a published legal page. TODO — OWNER INPUT REQUIRED: final publication URL.

1. Service description and scope

agentSAFE performs AI-assisted screening of Ethereum Mainnet smart contracts. For a contract address you provide, the Service retrieves the verified source code, runs third-party static and symbolic analysis tools, and uses large language model ("LLM") agents to summarize potential risks. The output is a machine-readable report of findings for human review.

The Service screens contracts. It does not certify, approve, or endorse any contract, and it is not a substitute for a professional security audit.

2. AI-assisted analysis limitations

Findings are generated with the assistance of LLM agents and automated tools. LLMs can produce incorrect, incomplete, or misleading output ("hallucination"), and automated tools have known blind spots. Analysis quality depends on the availability of verified source code and on-chain data at the time of the request. Results represent a point-in-time assessment only.

3. No guarantee of vulnerability detection

The Service does not guarantee that all vulnerabilities will be detected. Absence of findings is not evidence of safety.

4. No guarantee of contract safety

A report with few or no findings does not mean the contract is safe, and must not be represented as such to any third party.

5. User responsibility

You are solely responsible for verifying all findings, for any action you take based on a report, and for any decision to deploy, use, or interact with the audited contract. TODO — OWNER INPUT REQUIRED: confirm this allocation of responsibility with counsel.

6. Pricing, payment authorization, and service fees

Prices are denominated in USDC and settled via the x402 payment protocol on Base Mainnet (eip155:8453) through the PayAI facilitator (or the network advertised by the Service at request time). Prices are tiered by verified source-code size and are advertised by the Service at request time, before payment authorization:

Source lines are counted by the Service from the verified source published by the block explorer; blank lines are excluded and the count is determined solely by the Service, never by a client-submitted value. For contracts exceeding 5,000 source lines, the Service rejects the request before payment with the error code ENTERPRISE_INQUIRY_REQUIRED and an inquiry link: https://tally.so/r/laby7B.

Calling a paid tool without payment returns a payment request; your payment application authorizes the exact advertised amount before the Service executes your request. Third-party payment infrastructure (facilitators, blockchain networks) may charge its own settlement and network fees, which are separate from the Service price.

7. Audit execution failures, partial results, and report availability

The Service runs each audit on ephemeral infrastructure with a bounded execution time. If one or more analysis tools fail, the report is marked as a partial audit and lists the tools that failed; results from tools that completed are preserved. A partial audit is never presented as a fully completed one, and a failed tool is never presented as a successful scan with zero findings. Reports are retained for a limited period (currently 2 days, per the report retention setting); retrieve your report promptly after completion.

8. Refund and failed-service handling

AI audit reports are virtual digital products generated instantly. Once execution has completed successfully and the report has been generated and delivered, the purchase is non-refundable. If a technical failure results in an empty or missing report, the user is entitled to a full refund upon contacting support at lunanoach@gmail.com.

For clarity, consistent with the payment flow: a payment is settled only after the requested work has executed successfully, and a failed execution never settles — in that case no payment is collected and no refund is owed. The full refund entitlement above applies when a payment settled but the report is empty or missing.

Mandatory consumer-protection requirements applicable in Hong Kong or the user's jurisdiction may override this section; this policy is drafted for legal review and is not presented as guaranteed enforceable in every circumstance. TODO — OWNER INPUT REQUIRED: confirm with counsel that this refund rule satisfies applicable mandatory consumer-protection law.

9. Report access tokens and confidentiality

Audit reports are associated with an unguessable task identifier issued at request time. Keep your task identifiers confidential: possession of a valid identifier is the authorization used to retrieve the corresponding report. The Service does not publish your reports. The Service is not a secret store, however — treat report contents as non-confidential data.

10. Data handling, source-code processing, and retention

The Service processes only: (a) the contract address you submit, (b) public verified source code and on-chain data it retrieves for that address, (c) payment metadata required by the x402 protocol, and (d) the unguessable task identifier issued at request time. Reports and task records are retained for the period stated in Section 7 and then expire. The Service does not knowingly process personal data. See the separate PRIVACY_POLICY.md, which describes the actual data flows, including third parties (payment facilitator, block explorer, RPC provider, code-hosting, LLM providers) and transport-level logging. TODO — OWNER INPUT REQUIRED: privacy practices must be reviewed for applicable Hong Kong requirements (e.g. the Personal Data (Privacy) Ordinance) before publication.

11. Acceptable use and prohibited activity

You agree not to: attempt to bypass payment or replay payment payloads; reverse-engineer the Service to resell its output as your own security certification; submit malicious automated workloads or attempt to disrupt the Service; use the Service in violation of applicable law or sanctions; or misrepresent Service output to third parties.

12. Service availability and maintenance

The Service is offered "as available". It may be paused for maintenance, or degraded by third-party outages (RPC providers, explorers, GitHub Actions, LLM providers, payment facilitators). There is no uptime guarantee. TODO — OWNER INPUT REQUIRED: state any intended SLA, or explicitly state that none is offered.

13. Limitation of liability

To the maximum extent permitted by law, the Service is provided "as is" and "as available" without warranties of any kind, express or implied, including merchantability, fitness for a particular purpose, and non-infringement. The operator is not liable for any damages, losses, or costs arising from use of, or reliance on, the Service or its reports. This section is drafted for legal review and is not presented as guaranteed enforceable. Governing law and jurisdiction: Hong Kong, as confirmed by the operator. TODO — OWNER INPUT REQUIRED: confirm any liability cap with counsel.

14. Contact

Operator / legal entity: workfern (Hong Kong).

Contact e-mail: lunanoach@gmail.com

Commercial inquiries, refunds, and enterprise/custom pricing: https://tally.so/r/laby7B