Back to skill

Security audit

Agnic

Security checks for vulnerabilities and agentic risk

Overview

This skill openly provides wallet, payment, trading, email, and identity powers, but it uses an unpinned remote command package with broad command access and lacks explicit safeguards for irreversible actions.

Review carefully before installing. Use only if you trust the agnic npm package and publisher, and prefer a pinned, reviewed version with narrow command permissions. Require explicit confirmation before any payment, trade, token transfer, email send, or email reply, and avoid exposing unrelated credentials or private files to the CLI environment.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (1)

T08 · Insecure Dependencies

Error
Location
SKILL.md:8
Finding

Execution of an Unpinned, Mutable npm Package with Broad Tool Authorization

Content
View full analysis
--method GET --json npx agnic@latest trade 10 USDC ETH --json npx agnic@latest send --network base --json npx agnic@latest email send --to user@example.com --subject "Hello" --body "Message" ``` ### Technical Analysis The skill authorizes execution of `npx agnic@latest`, which resolves a mutable package version from the npm ecosystem at invocation time. The reviewed project contains no pinned package version, lockfile, integrity hash, vendored source, or trusted wrapper that would ensure the executed code is the same code that was reviewed. Consequently, publishing a new `latest` release—or compromising the package publisher, registry account, or distribution path—can alter the code executed by the skill without any modification to `SKILL.md`. Because `npx` executes package-provided code locally, a malicious dependency release could run with the operating-system permissions of the agent process. The trailing wildcard in: ```yaml "Bash(npx agnic@latest *)" ``` also authorizes arbitrary arguments and package subcommands rather than limiting access to the documented operations. This increases exposure if the package has undocumented, newly added, or dangerous subcommands. The issue is especially significant because the skill uses the dependency for security-sensitive capabilities: OAuth authentication, stored account access, wallet balances, paid API calls, mainnet token trades, USDC transfers, identity data, email, and AI prompts. The package implementation itself ...[truncated 2241 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (30)

Rp1

Medium
Category
MCP Rug Pull
Confidence
99% confidence
Finding

The manifest itself authorizes Bash(npx agnic@latest *), meaning the agent can execute arbitrary subcommands from a mutable external package. Because the skill includes wallet, trading, payment, email, and identity actions, this creates a critical remote-code and supply-chain risk with direct access to high-value operations.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The invocation description is broad enough to match many common wallet or email requests without clear scope boundaries. For a skill capable of payments, trades, messaging, and identity actions, overbroad triggering increases the risk of activating powerful tooling in contexts where the user did not intend such access.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

The manifest allows execution via Bash(npx agnic@latest *), which fetches and runs code at invocation time from a moving package target. In a wallet/payment/trading skill, this creates a supply-chain risk where a compromised or malicious package update could immediately gain access to sensitive actions and credentials.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

Documenting npx agnic@latest auth login reinforces use of an unpinned remote package for authentication flows. If the package is replaced upstream, users may be directed through a malicious OAuth flow or credential-harvesting logic.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

Using npx agnic@latest auth logout still relies on executing whatever code the registry currently serves. In this context, even seemingly harmless auth commands can be abused to tamper with stored credentials or session state.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

Balance queries are routed through an unpinned package, so a registry compromise could alter outputs, misreport balances, or exfiltrate wallet metadata. In a financial skill, false balances can influence later irreversible actions.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
96% confidence
Finding

The network-specific balance command also executes unpinned code from the package registry. Because the skill handles assets across networks, this expands the blast radius of a supply-chain compromise to multiple environments.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

Even address display commands run unpinned remote code, which can leak identifiers or substitute attacker-controlled addresses in outputs. In a wallet context, that can directly cause misdirected transfers if users trust the shown address.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The documentation presents irreversible financial and messaging commands—payments, trades, sends, and email send/reply—without explicit warning, confirmation, or review guidance. In this context, omission of safeguards is dangerous because users or agents may execute high-impact actions based on incomplete understanding or manipulated inputs.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The x402 search command uses an unpinned package and may return attacker-influenced endpoints or manipulated service information. Because it feeds into later payment actions, tampering here can steer users toward malicious paid APIs.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

Previewing payment costs via unpinned code allows an upstream compromise to falsify fee estimates or destination details. That increases the chance of users authorizing unexpected charges.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

x402 pay performs a payment action through code fetched at runtime with @latest. In a direct payment path, a supply-chain takeover could alter recipient, amount, request content, or exfiltrate auth/payment material.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

Trade quote retrieval through an unpinned package can manipulate price quotes, slippage expectations, or route selection. Misleading quote data in a trading workflow can cause financial loss even before execution.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
99% confidence
Finding

Executing trades through npx agnic@latest trade is especially dangerous because it combines runtime code fetching with irreversible asset movement. A malicious upstream release could substitute assets, alter trade amounts, or drain funds.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
99% confidence
Finding

The send command enables direct USDC transfers using unpinned registry code. In a wallet skill, this is a critical supply-chain exposure because any upstream compromise can immediately redirect funds or initiate unauthorized transfers.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

Identity queries executed through unpinned code could leak identity metadata or falsify trust/delegation information. In agent identity workflows, manipulated outputs may affect access decisions or trust judgments.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

General account status is retrieved via unpinned code, allowing a compromised package to misrepresent authentication, balances, or account readiness. This can prime users to take unsafe follow-up actions.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

Displaying the email alias through unpinned code can expose or substitute addresses, leading to phishing or misdirected communication. In a skill that handles identity and payments, messaging integrity matters.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

Email setup via an unpinned package could create aliases under attacker control or capture setup metadata. Because this becomes part of the agent's identity surface, compromise can persist beyond a single run.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
94% confidence
Finding

Inbox access through unpinned code can exfiltrate message contents, tokens, and verification links. Since the documented workflow includes sign-up and verification, mailbox compromise can lead to broader account takeover.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

Sending email through remotely fetched unpinned code risks unauthorized outbound messages, data leakage, or phishing from the agent identity. This can damage trust and be used to advance fraud or account compromise.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

Replying to email with unpinned code combines access to message context with outbound transmission under agent identity. A compromised package could leak thread contents or send manipulated replies that authorize or verify attacker-controlled actions.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill is described as a wallet/payment/trading/email/identity tool, but it also exposes general AI chat and image generation capabilities. This scope drift increases the chance that sensitive wallet or inbox context is routed into unrelated model or file-generation flows, violating least privilege.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The undocumented-for-purpose AI gateway features are not justified by the skill's stated financial and identity mission, creating unnecessary data egress and command surface. In a high-privilege skill, unrelated capabilities materially raise the risk of misuse or accidental sensitive-data exposure.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
93% confidence
Finding

Model listing via unpinned code is less critical than funds movement, but still permits registry-delivered code execution and possible telemetry leakage. In a skill already handling sensitive data, any extra execution surface increases risk.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.