Back to skill

Security audit

Transact

Security checks for vulnerabilities and agentic risk

Overview

This wallet assistant requires careful transaction review, but its setup and network instructions exceed the permissions it declares.

Before installing, reconcile the declared permissions with the required RPC hosts, installation location, export directory, and persistent MCP configuration. Use a reviewed CLI version and keep signing under your control. Aomi receives your transaction request and wallet context; RPC providers receive balance and transaction queries. The artifacts support Review for these scope inconsistencies, not a malicious classification.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (13)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied commands perform general social-media information retrieval, not crypto, DeFi, or wallet execution. The description does not declare X integration or these social-media capabilities, and the examples are not presented as supporting any trading workflow. Although the commands use the permitted aomi CLI, that does not make their functionality consistent with the declared purpose. The chunk does not establish a network-permission violation or credential handling issue.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The hosted MCP instructions invoke codex and persist MCP configuration despite authorizing only Aomi-related executables and writes under ~/.aomi. If the host does not enforce those restrictions, following the instructions expands persistent tool configuration outside the declared authority; a strict host would instead block setup.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The commands perform social-network user discovery rather than the declared chain-state or crypto/DeFi operations. Although they use the permitted Aomi CLI, the requested capability is outside the stated purpose. The chunk does not establish direct access to an unauthorized network destination or any credential handling.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The transaction workflow matches the declared crypto-agent purpose. However, aomi tx export <commit-id> --format commit > reviewed-commit.json uses shell redirection to create or overwrite a file in the current directory. Nothing in the supplied chunk confines that directory to ~/.aomi/, so this command permits a write outside the declared filesystem boundary. No other mismatch is evident from the supplied commands.

Content

No source excerpt is available for this finding.

Manifest references MCP server without version pin: 'npx @aomi-labs/cli'.

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

The manifest authorizes a mutable @latest package rather than fixing the executable dependency to a reviewed release. This duplicates the other version-pinning finding, but independently identifies a genuine supply-chain exposure in a workflow handling financial transactions and local credentials.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx @aomi-labs/cli'.

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

The shell declaration uses @latest rather than an immutable CLI version, so subsequent execution can fetch code that was not reviewed with this skill. Although this is a CLI rather than an MCP server, the supply-chain risk is real and particularly consequential because the CLI can access wallet signing material.

Content

No source excerpt is available for this finding.

The preparation workflow sends transaction intent and wallet information to Aomi without an explicit privacy disclosure.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

The workflow instructs the agent to submit the user's exact intent, sender public key, and chain ID through Aomi's backend-backed chat service. Although the file identifies the hosted service, it does not warn that these potentially sensitive financial details leave the local environment and are shared with Aomi.

Content

No source excerpt is available for this finding.

Hosted MCP setup requires commands outside the declared shell permissions.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The hosted setup explicitly calls codex, which is not in the declared shell permissions. This is the same authorization-boundary issue identified by the configuration mismatch finding, not evidence of a malicious endpoint or a demonstrated sandbox bypass.

Content

No source excerpt is available for this finding.

The installation instructions require a global npm installation outside the declared shell and filesystem permissions.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The instructions invoke npm, which is not an authorized shell command, and install into a global prefix outside the declared ~/.aomi write scope. If permissions are advisory, this can persist executable changes in shared tooling locations; an enforcing host should reject it.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx @aomi-labs/cli'.

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

The one-shot execution example resolves @latest, allowing an unreviewed future package to run with the agent's privileges. The stated minimum version and capability checks do not authenticate the code, and a compromised CLI could expose signing material or alter transaction handling.

Content

No source excerpt is available for this finding.

The Commit export command writes to the current directory rather than the permitted state directory.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The shell redirection writes reviewed-commit.json relative to an unspecified working directory, which need not be within ~/.aomi. It can truncate an existing file or place transaction metadata in an unintended shared or version-controlled location when filesystem restrictions are not enforced.

Content

No source excerpt is available for this finding.

Troubleshooting directs RPC access beyond the permitted network host.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

Receipt inspection on a chain-matching RPC requires network access not covered by the sole allowed host, chat.aomi.dev. The activity is legitimate for reconciliation, but the instructions neither obtain an additional grant nor constrain the RPC endpoint, creating an authorization mismatch and potential disclosure of wallet activity to another service.

Content

No source excerpt is available for this finding.

The funding workflow requires RPC connections outside the manifest's network allowlist.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The workflow explicitly requires direct RPC requests and provides publicnode.com endpoints, neither of which is covered by the chat.aomi.dev-only network permission. These are ordinary balance checks rather than demonstrated malicious exfiltration, but executing them without an additional grant violates the declared network boundary and discloses queried wallet addresses to third-party providers.

Content

No source excerpt is available for this finding.