Back to skill

Security audit

Trade Hunter

Security checks for vulnerabilities and agentic risk

Overview

This lead-generation skill is mostly coherent, but it collects business contact data with weak privacy framing and has an open-ended install dependency.

Review this before installing if you will use it for outreach. Treat returned contact emails as potentially regulated personal or business data, confirm your legal basis and regional marketing rules, and avoid automatic contact collection unless the user explicitly asks for it. Pin and review the coze-api dependency before deployment.

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

Warning
Location
package.json:22
Finding

Unbounded Third-Party Dependency Without Lockfile Integrity Pinning

Content
View full analysis

Vulnerability Details

File Location: package.json, lines 22–23
Vulnerability Type: Unbounded dependency version range and mutable supply-chain resolution
Risk Level: Medium

Complete Code Snippet:

json
"dependencies": {
  "coze-api": ">=1.0.0"
}

Technical Analysis

The dependency constraint accepts every coze-api release from version 1.0.0 onward, including future major versions. The project also contains no package lockfile or integrity-pinned dependency resolution. As a result, installing the same project at different times can retrieve materially different third-party code that was not part of this audit.

Package installation may execute dependency lifecycle scripts, while imported dependency code executes with the privileges of the application process. If the package publisher account or distribution channel is compromised, or if a future accepted release is malicious, an installation can introduce attacker-controlled code without any change to this repository. The broad range also permits unreviewed breaking changes that could weaken security assumptions.

Attack Path

  1. An attacker compromises the coze-api publishing account or otherwise causes a malicious version greater than or equal to 1.0.0 to be distributed through the configured package registry.
  2. A user or automated build installs this project without a pre-existing integrity-pinned lockfile.
  3. The package manager resolves the open-ended range to the attacker-controlled release.
  4. Malicious lifecycle scripts may execute during installation, or malicious module logic may execute when the dependency is loaded at runtime.
  5. The payload operates with the permissions and environmental access of the installer, build worker, or application process.

This path requires compromise or malicious control of an accepted dependency release; the audited repository itself does not contain a malicious payload.

Impact

...[truncated 540 chars]

Remediation
View remediation

Remediation Suggestions

  1. Replace the open-ended range with an exact version that has been reviewed, for example "coze-api": "1.0.0" or another specifically audited release.
  2. Generate and commit the package manager's lockfile so transitive versions and registry integrity hashes are reproducible.
  3. Use deterministic installation commands such as npm ci in CI/CD pipelines, and fail builds when the lockfile and manifest differ.
  4. Review the dependency's ownership, release history, lifecycle scripts, provenance, and transitive dependency tree before adoption.
  5. Disable package lifecycle scripts where operationally feasible, such as with npm ci --ignore-scripts, and explicitly allow only scripts that are required and reviewed.
  6. Run dependency installation and application execution with least privilege, isolated credentials, restricted filesystem access, and constrained outbound network access.
  7. Apply dependency updates through a controlled review process with automated vulnerability scanning and tests rather than accepting future releases automatically.
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (7)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill title, descriptions, usage scenarios, scoring dimensions, and related-skill labels are all presented in Chinese, with no indication that users may choose another language or that the skill is intentionally limited to a Chinese-speaking audience. Under the language/locale policy, a skill should not impose a specific language by default without opt-in or a clearly documented justification.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The trigger conditions are broad enough to match many ordinary business-search requests, which increases the chance of unintended invocation. Because this skill performs external data collection and may return contact information, accidental activation can expose users to privacy, compliance, and data-quality risks without clear intent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill advertises collection of contact information but does not present meaningful warnings about privacy obligations, data accuracy, consent, or jurisdiction-specific compliance requirements. In a lead-generation context, that omission is risky because users may rely on the tool for outreach workflows that implicate personal data and marketing regulations.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The documentation states that PII has been desensitized, but the example output explicitly includes a direct contact email field. That contradiction can mislead users and downstream agents into handling personal or business contact data as if it were already sanitized, increasing privacy, compliance, and misuse risk.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

The description, headings, trigger conditions, examples, and notes are all written in Chinese, which effectively forces a language choice for users reading the skill documentation. There is no opt-in language selection or explicit statement that the skill is intentionally restricted to Chinese-speaking users or a specific region.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
79% confidence
Finding

The CSV header row uses Chinese field names, which imposes a specific language on consumers of the file. There is no accompanying indication that the locale is optional or that the file is intentionally region-specific, so this may violate language/locale policy expectations.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
94% confidence
Finding

The package description is entirely in Chinese, and the manifest does not indicate that the skill is region-specific or that users can choose another language. Under the policy for natural-language issues, this can be treated as a locale/language constraint without explicit opt-in or justification.

Content

No source excerpt is available for this finding.