Management Summary
Coti contacted Sayfer Security in order to perform penetration testing on Coti’s MetaMask Snap in 03/2026.
Before assessing the above services, we held a kickoff meeting with the Coti technical team and received an overview of the system and the goals for this research.
Over the research period of 2 weeks, we discovered 7 vulnerabilities in the system.
In conclusion, several fi xes should be implemented following the report, but the system’s security posture is competent.
After a review by the Sayfer team, we certify that all the security issues mentioned in this report have been addressed by the Coti team.
Risk Methodology
At Sayfer, we are committed to delivering the highest quality penetration testing to our clients. That’s why we have implemented a comprehensive risk assessment model to evaluate the severity of our fi ndings and provide our clients with the best possible recommendations for mitigation.
Our risk assessment model is based on two key factors: IMPACT and LIKELIHOOD. Impact refers to the potential harm that could result from an issue, such as fi nancial loss, reputational damage, or a non-operational system. Likelihood refers to the probability that an issue will occur, taking into account factors such as the complexity of the attack and the number of potential attackers.
By combining these two factors, we can create a comprehensive understanding of the risk posed by a particular issue and provide our clients with a clear and actionable assessment of the severity of the issue. This approach allows us to prioritize our recommendations and ensure that our clients receive the best possible advice on how to protect their business.
Risk is defined as follows:

Vulnerabilities by Risk
High – Direct threat to key business processes.
Medium – Indirect threat to key business processes or partial threat to business processes.
Low – No direct threat exists. The vulnerability may be exploited using other vulnerabilities.
Informational – This finding does not indicate vulnerability, but states a comment that notifies about design flaws and improper implementation that might cause a problem in the long run.
| Severity | # of issues |
|---|---|
| Low | 3 |
| Medium | 4 |
| High | 0 |
| Informational | 0 |
Approach
Introduction
Coti contacted Sayfer to perform penetration testing on their MetaMask Snap application.
This report documents the research carried out by Sayfer targeting the selected resources defi ned under the research scope. Particularly, this report displays the security posture review for Coti’s MetaMask Snap application and its surrounding infrastructure and process implementations.
Our penetration testing project life cycle:
Scope Overview
Technical Overview
Scope Validation
Threat Model
Security Evaluation
Security Assessment
Scope Overview
During our fi rst meeting and after understanding the company’s needs, we defi ned the application’s scope that resides at the following URLs as the scope of the project:
- Coti’s MetaMask Snap
- Audit commit: ac5d55a1edd011db0c214d2f83bb0d811e35305f
- Fixes commit: 39ac8e58a4238401371ca798f4371732b8ec5c18
Our tests were performed from 12/03/2026 to 26/03/2026.
Scope Validation
We began by ensuring that the scope defi ned to us by the client was technically logical.
Deciding what scope is right for a given system is part of the initial discussion. Getting the scope right is key to deriving maximum business value from the research.
Threat Model
During our kickoff meetings with the client we defi ned the most important assets the application possesses.
We defi ned that the largest current threat to the system is leakage of the user’s AES encryption key, which protects confi dential balances and underpins wallet key derivation, leading to potential loss of funds and full compromise of confi dential transaction data.
Security Evaluation Methodology
Sayfer uses OWASP WSTG as our technical standard when reviewing web applications. After gaining a thorough understanding of the system we decided which OWASP tests are required to evaluate the system.
Security Assessment
After understanding and defi ning the scope, performing threat modeling, and evaluating the correct tests required in order to fully check the application for security fl aws, we performed our security assessment.
Issue Table Description
Issue title
| ID | SAY-??: An ID for easy communication on each vulnerability |
| Status | Open/Fixed/Acknowledged |
| Risk | Represents the risk factor of the issue. For further description refer to the Vulnerabilities by Risk section. |
| Business Impact | The main risk of the vulnerability at a business level. |
| Location | The URL or the fi le in which this issue was detected. Issues with no location have no particular location and refer to the product as a whole. |
| Description | Here we provide a brief description of the issue and how it formed, the steps we made to fi nd or exploit it, along with proof of concept (if present), and how this issue can aff ect the product or its users. |
| Mitigation | Suggested resolving options for this issue and links to advised sites for further remediation. |
Security Evaluation
The following tests were conducted while auditing the system
Information Gathering
| Information Gathering | Test Name | Status |
| WSTG-INFO-01 | Conduct Search Engine Discovery Reconnaissance for Information Leakage | Pass |
| WSTG-INFO-02 | Fingerprint Web Server | Pass |
| WSTG-INFO-03 | Review Webserver Metafiles for Information Leakage | Pass |
| WSTG-INFO-04 | Enumerate Applications on Webserver | Pass |
| WSTG-INFO-05 | Review Webpage Content for Information Leakage | Pass |
| WSTG-INFO-06 | Identify application entry points | Pass |
| WSTG-INFO-07 | Map execution paths through application | Pass |
| WSTG-INFO-08 | Fingerprint Web Application Framework | Pass |
| WSTG-INFO-09 | Fingerprint Web Application | Pass |
| WSTG-INFO-10 | Map Application Architecture | Pass |
Configuration and Deploy Management Testing
| Configuration and Deploy Management Testing | Test Name | Status |
| WSTG-CONF-01 | Test Network Infrastructure Configuration | Pass |
| WSTG-CONF-02 | Test Application Platform Configuration | Pass |
| WSTG-CONF-03 | Test File Extensions Handling for Sensitive Information | Pass |
| WSTG-CONF-04 | Review Old Backup and Unreferenced Files for Sensitive Information | Pass |
| WSTG-CONF-05 | Enumerate Infrastructure and Application Admin Interfaces | Pass |
| WSTG-CONF-06 | Test HTTP Methods | Pass |
| WSTG-CONF-07 | Test HTTP Strict Transport Security | Pass |
| WSTG-CONF-08 | Test RIA cross domain policy | Pass |
| WSTG-CONF-09 | Test File Permission | Pass |
| WSTG-CONF-10 | Test for Subdomain Takeover | Pass |
| WSTG-CONF-11 | Test Cloud Storage | Pass |
Identity Management Testing
| Identity Management Testing | Test Name | Status |
| WSTG-IDNT-01 | Test Role Definitions | Pass |
| WSTG-IDNT-02 | Test User Registration Process | Pass |
| WSTG-IDNT-03 | Test Account Provisioning Process | Pass |
| WSTG-IDNT-04 | Testing for Account Enumeration and Guessable User Account | Pass |
| WSTG-IDNT-05 | Testing for Weak or unenforced username policy | Pass |
Authentication Testing
| Authentication Testing | Test Name | Status |
| WSTG-ATHN-01 | Testing for Credentials Transported over an Encrypted Channel | Pass |
| WSTG-ATHN-02 | Testing for Default Credentials | Pass |
| WSTG-ATHN-03 | Testing for Weak Lock Out Mechanism | Pass |
| WSTG-ATHN-04 | Testing for Bypassing Authentication Schema | Pass |
| WSTG-ATHN-05 | Testing for Vulnerable Remember Password | Pass |
| WSTG-ATHN-06 | Testing for Browser Cache Weaknesses | Pass |
| WSTG-ATHN-07 | Testing for Weak Password Policy | Pass |
| WSTG-ATHN-08 | Testing for Weak Security Question Answer | Pass |
| WSTG-ATHN-09 | Testing for Weak Password Change or Reset Functionalities | Pass |
| WSTG-ATHN-10 | Testing for Weaker Authentication in Alternative Channel | Pass |
Authorization Testing
| Authorization Testing | Test Name | Status |
| WSTG-ATHZ-01 | Testing Directory Traversal File Include | Pass |
| WSTG-ATHZ-02 | Testing for Bypassing Authorization Schema | Pass |
| WSTG-ATHZ-03 | Testing for Privilege Escalation | Pass |
| WSTG-ATHZ-04 | Testing for Insecure Direct Object References | Pass |
Session Management Testing
| Session Management Testing | Test Name | Status |
| WSTG-SESS-01 | Testing for Session Management Schema | Pass |
| WSTG-SESS-02 | Testing for Cookies Attributes | Pass |
| WSTG-SESS-03 | Testing for Session Fixation | Pass |
| WSTG-SESS-04 | Testing for Exposed Session Variables | Pass |
| WSTG-SESS-05 | Testing for Cross Site Request Forgery | Pass |
| WSTG-SESS-06 | Testing for Logout Functionality | Pass |
| WSTG-SESS-07 | Testing Session Timeout | Pass |
| WSTG-SESS-08 | Testing for Session Puzzling | Pass |
| WSTG-SESS-09 | Testing for Session Hijacking | Pass |
Data Validation Testing
| Data Validation Testing | Test Name | Status |
| WSTG-INPV-01 | Testing for Reflected Cross Site Scripting | Pass |
| WSTG-INPV-02 | Testing for Stored Cross Site Scripting | Pass |
| WSTG-INPV-03 | Testing for HTTP Verb Tampering | Pass |
| WSTG-INPV-04 | Testing for HTTP Parameter Pollution | Pass |
| WSTG-INPV-05 | Testing for SQL Injection | Pass |
| WSTG-INPV-06 | Testing for LDAP Injection | Pass |
| WSTG-INPV-07 | Testing for XML Injection | Pass |
| WSTG-INPV-08 | Testing for SSI Injection | Pass |
| WSTG-INPV-09 | Testing for XPath Injection | Pass |
| WSTG-INPV-10 | Testing for IMAP SMTP Injection | Pass |
| WSTG-INPV-11 | Testing for Code Injection | Pass |
| WSTG-INPV-12 | Testing for Command Injection | Pass |
| WSTG-INPV-13 | Testing for Format String Injection | Pass |
| WSTG-INPV-14 | Testing for Incubated Vulnerability | Pass |
| WSTG-INPV-15 | Testing for HTTP Splitting Smuggling | Pass |
| WSTG-INPV-16 | Testing for HTTP Incoming Requests | Pass |
| WSTG-INPV-17 | Testing for Host Header Injection | Pass |
| WSTG-INPV-18 | Testing for Server-side Template Injection | Pass |
| WSTG-INPV-19 | Testing for Server-Side Request Forgery | Pass |
Error Handling
| Error Handling | Test Name | Status |
| WSTG-ERRH-01 | Testing for Improper Error Handling | Pass |
| WSTG-ERRH-02 | Testing for Stack Traces | Pass |
Cryptography
| Cryptography | Test Name | Status |
| WSTG-CRYP-01 | Testing for Weak Transport Layer Security | Pass |
| WSTG-CRYP-02 | Testing for Padding Oracle | Pass |
| WSTG-CRYP-03 | Testing for Sensitive Information Sent via Unencrypted Channels | Pass |
| WSTG-CRYP-04 | Testing for Weak Encryption | Pass |
Business logic Testing
| Business Logic Testing | Test Name | Status |
| WSTG-BUSL-01 | Test Business Logic Data Validation | Pass |
| WSTG-BUSL-02 | Test Ability to Forge Requests | Pass |
| WSTG-BUSL-03 | Test Integrity Checks | Pass |
| WSTG-BUSL-04 | Test for Process Timing | Pass |
| WSTG-BUSL-05 | Test Number of Times a Function Can be Used Limits | Pass |
| WSTG-BUSL-06 | Testing for the Circumvention of Work Flows | Pass |
| WSTG-BUSL-07 | Test Defenses Against Application Mis-use | Pass |
| WSTG-BUSL-08 | Test Upload of Unexpected File Types | Pass |
| WSTG-BUSL-09 | Test Upload of Malicious Files | Pass |
Client Side Testing
| Client Side Testing | Test Name | Status |
| WSTG-CLNT-01 | Testing for DOM-Based Cross Site Scripting | Pass |
| WSTG-CLNT-02 | Testing for JavaScript Execution | Pass |
| WSTG-CLNT-03 | Testing for HTML Injection | Pass |
| WSTG-CLNT-04 | Testing for Client Side URL Redirect | Pass |
| WSTG-CLNT-05 | Testing for CSS Injection | Pass |
| WSTG-CLNT-06 | Testing for Client Side Resource Manipulation | Pass |
| WSTG-CLNT-07 | Test Cross Origin Resource Sharing | Pass |
| WSTG-CLNT-08 | Testing for Cross Site Flashing | Pass |
| WSTG-CLNT-09 | Testing for Clickjacking | Pass |
| WSTG-CLNT-10 | Testing WebSockets | Pass |
| WSTG-CLNT-11 | Test Web Messaging | Pass |
| WSTG-CLNT-12 | Testing Browser Storage | Pass |
| WSTG-CLNT-13 | Testing for Cross Site Script Inclusion | Pass |
API Testing
| API Testing | Test Name | Status |
| WSTG-APIT-01 | Testing GraphQL | Pass |
Security Assessment Findings
No Confirmation Dialogue in set-aes-key
| ID | SAY-01 |
| Status | Fixed |
| Risk | Medium |
| Business Impact | A malicious or compromised dapp with RPC access can silently replace the user’s encryption key with an attacker-controlled one, causing all subsequent confi dential token transfers to be encrypted with the attacker’s key. The attacker can then decrypt all confi dential balances and transaction amounts, while the user loses the ability to view their own balances. |
| Location | – ../src/index.tsx:817 |
Description
The set-aes-key RPC handler directly writes the provided newUserAesKey to state without any user-facing confirmation dialog. This stands in contrast with other sensitive RPC handlers in the same fi le: delete-aes-key requires a full snap_dialog confirmation and get-aes-key also prompts the user for approval before returning key material.
Additionally, the snap manifest declares "endowment:rpc": { "dapps": true }, meaning any dapp origin can invoke this method, i.e. there is no whitelist of trusted origins. The handler also accepts any non-empty string as a valid AES key without checking hex format or length, making it possible to set weak or malformed keys.
Mitigation
- Add a snap_dialog confirmation before writing the AES key, clearly showing the requesting origin and a warning about key replacement.
- Validate the AES key format (must be exactly 32 hex characters / 16 bytes for AES-128) before storing.
- Consider restricting the allowed origins for this sensitive operation to only the companion dapp or alternatively to a whitelist of approved dapps.
Consider restricting the allowed origins for this sensitive operation to only the companion dapp or alternatively to a whitelist of approved dapps.
| ID | SAY-02 |
| Status | Fixed |
| Risk | Medium |
| Business Impact | Similarly to SAY-01, any dapp can silently add or remove tokens from the user’s snap token list without user consent. A malicious dapp can fl ood the list with phishing tokens that mimic legitimate assets to trick users into interacting with malicious contracts, or silently remove legitimate tokens from the user’s view. |
| Location | – …/src/index.tsx:945, 1006 |
Description
The import-token handler accepts arbitrary token addresses, names, symbols, and decimals from any calling dapp and writes them directly to the snap’s persistent state with no user confirmation dialog. Similarly, hide-token removes tokens by address without any confirmation.
Both handlers completely bypass the user. This is inconsistent with the snap’s own internal UI fl ow for importing ERC20/ERC721 tokens via onUserInput, which requires user interaction through form validation and UI confirmation steps.
Since the snap manifest grants RPC access to all dapps, any website can invoke these methods.
Mitigation
Add snap_dialog confirmation for both import-token and hide-token operations, displaying the token details (address, name, symbol) and the requesting origin before modifying the token list.
Unchecked ERC20 Transfer Return Value
| ID | SAY-03 |
| Status | Fixed |
| Risk | Medium |
| Business Impact | While the handler does prompt for user approval, the confirmation dialog uses vague wording that does not communicate that raw cryptographic key material will be sent to the requesting website. Once a dapp obtains the raw AES key, it can decrypt all confi dential balances and craft encrypted transfer amounts offl ine. |
| Location | – …/src/index.tsx:724 |
Description
get-aes-key returns the raw value of currentAESState.aesKey to the calling dapp after the user approves a dialog. The confirmation message reads “Approve to unlock your security key”, but does not explicitly warn that the raw cryptographic key will be exposed to the website.
Users may approve this dialog thinking it is a benign authentication or unlock operation, not realizing they are handing over the master encryption key that protects all their confi dential balances. The generic wording provides inadequate informed consent for such a sensitive operation.
Any dapp can trigger this dialog. There is no restriction on which origins may request the AES key, meaning phishing sites or compromised dapps can obtain it if the user clicks “Approve”.
Mitigation
Rewrite the confirmation dialog to be explicit – “This will share your raw AES encryption key with {origin}. This key can decrypt all your confi dential balances and transactions.”
Arbitrary Digest Signing without Meaningful Context
| ID | SAY-04 |
| Status | Fixed |
| Risk | Medium |
| Business Impact | Any dapp can request the snap to sign an arbitrary 32-byte digest using the snap-derived private key. The confirmation dialog only shows raw hex, which is uninterpretable to users. This could be exploited to forge signatures for COTI MPC precompile operations or other on-chain actions tied to the snap’s derived wallet address. |
| Location | – …/src/index.tsx:656 |
Description
The sign-raw-256 handler signs any 32-byte hex value using buildRawSignature(), which derives a wallet from deriveSnapWallet(aesKey, address, chainId). The only validation performed is isHex32(), a check that the input is exactly 32 bytes of hex data.
The user sees a dialog displaying the raw hex digest, which is completely opaque to human interpretation. Consequently, they have no way of knowing whether they are signing a benign operation or authorizing a malicious transaction. This is analogous to the well-known eth_sign blind-signing vulnerability in MetaMask, which MetaMask itself deprecated for exactly this reason.
Mitigation
- Add structured contextual data to the signing dialog explaining what is being signed.
- Consider restricting the handler to known function selectors or requiring additional context parameters.
SSRF via NFT Metadata
| ID | SAY-05 |
| Status | Fixed |
| Risk | Low |
| Business Impact | A malicious token contract can return a crafted tokenURI pointing to internal network addresses or cloud metadata endpoints, allowing server-side request forgery from the MetaMask extension context. |
| Location | – …/src/utils/image.ts:54 |
Description
The snap requests the endowment:network-access permission and fetches URLs derived from on-chain NFT token URI metadata. In fetchWithGatewayFallback, non-IPFS URLs are fetched directly via fetchWithTimeout(string, ?number) with no URL scheme validation, host validation, or IP address fi ltering.
The URI value originates from decrypted on-chain token metadata (getTokenURI(…) in token.ts), which is fully attacker-controlled if they deploy a malicious NFT contract and trick the user into importing it. The attacker can set the tokenURI to point to internal network resources like http://169.254.169.254/ for cloud metadata, or http://localhost:PORT/ for local services.
getSVGfromMetadata(string) then fetches the image fi eld from the returned JSON response, creating a second SSRF opportunity – the attacker controls both the initial URI and the URI within the metadata JSON.
Mitigation
Validate that fetched URLs use only https:// or ipfs:// schemes. Block private IP ranges like 10.0.0.0/8, 172.16.0.0/12, 192.168.0.0/16 and link-local addresses like 169.254.0.0/16.
Weak Address Validation in ERC20 Import Flow
| ID | SAY-06 |
| Status | Fixed |
| Risk | Low |
| Business Impact | The snap’s internal ERC20 import fl ow validates token addresses by string length only, not by Ethereum address format. Any 42-character string passes validation, potentially leading to confusing error messages when downstream operations fail. |
| Location | – …/src/index.tsx:380 |
Description
The ERC20 token import handler within onUserInput checks the submitted address with a simple length comparison, which only verifi es that the string is 42 characters long but does not validate hex format, the 0x prefi x, or EIP-55 checksum.
For example, a string like 0000000000000000000000000000000000000000ab (42 characters, no 0x prefi x) or 0xZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZ (42 characters, invalid hex) would pass this validation. The subsequent getERC20Details(string) call will fail when trying to query the address on-chain, but the error is generic rather than a clear validation message.
Mitigation
Use ethers.isAddress() for proper Ethereum address validation, which checks format, length, and optional EIP-55 checksum in a single call.
Unvalidated JSON.parse on Untrusted Input
| ID | SAY-07 |
| Status | Fixed |
| Risk | Low |
| Business Impact | A Malformed JSON will throw a SyntaxError that propagates as an unhandled RPC error, potentially leaking stack trace information to the calling dapp. |
| Location | – …/src/index.tsx:546 |
Description
In the decrypt RPC handler, JSON.parse(encryptedValue) is called directly on the value supplied by the calling dapp to extract ciphertext and r-fi elds. There is no prior validation that the parameter is a valid JSON or that the parsed result contains the expected fi elds before destructuring.
While the broader RPC switch statement may catch this error, the resulting error message will be a raw SyntaxError from JSON.parse like “Unexpected token X in JSON at position Y”, rather than a clean, context-appropriate error. This can leak internal implementation details to the calling dapp and provides a poor developer experience for legitimate integrators.
Mitigation
Wrap the JSON.parse call in a try-catch block and return a clean, descriptive error message.