Back to skill

Security audit

AI服装虚拟试穿效果图|AI-HIVE

Security checks for vulnerabilities and agentic risk

Overview

The skill mostly matches an AI-HIVE virtual try-on workflow, but it handles API credentials in a way that can send them to an arbitrary configured API endpoint.

Review before installing. Use only the default AI-HIVE endpoint, do not set AI_HIVE_BASE_URL or --base-url to an untrusted host, and check ~/.ai-hive/config.json before running. Treat uploaded person or garment images as sensitive, confirm authorization and cost before generation, and rotate the AI-HIVE key if it was ever used with an untrusted endpoint.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/imagegen.py:91
Finding

Bearer API Key Can Be Sent to an Attacker-Controlled Endpoint

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (11)

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The skill claims a concrete AI-HIVE/OpenAPI workflow including configuration lookup, reference upload, pricing snapshots, routing modes, async polling, and result download, but the described behavior is not actually implemented and appears to stop at planning output. This mismatch is dangerous because users or calling agents may rely on nonexistent safeguards, cost checks, authorization handling, or auditability controls, leading to unsafe execution assumptions, unexpected billing, or misuse of unauthorized media workflows.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
90% confidence
Finding

The skill advertises runnable scripts and API-driven generation workflows but does not declare any explicit tool/permission boundaries, even though it implies use of environment variables, filesystem access, network calls, and shell execution. In an agent environment, missing scope declarations can cause over-privileged execution or ambiguous trust boundaries, increasing the chance that the agent uses sensitive capabilities without informed approval or proper sandboxing.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The default prompt hard-codes Chinese output ('production-ready Chinese workflow'), which overrides normal user language expectations without explicit opt-in. This is primarily a prompt-safety and UX-control issue: it can cause unintended behavior, reduce transparency, and make downstream review harder for users who do not read Chinese, especially in a workflow that generates runnable commands and deliverables.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

Implicit invocation is enabled, but the file does not define narrow trigger constraints at the interface level; this increases the chance the skill will auto-activate in situations where the user did not clearly request it. Because this skill can generate executable AI-HIVE commands and production workflows for image-generation tasks, accidental invocation could lead to unintended processing, confusing outputs, or unauthorized use of sensitive photos/workflows.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
82% confidence
Finding

The file’s user-facing title and CLI description are entirely in Chinese, which implies a fixed language experience without any opt-in or alternate locale support. The policy specifically calls for flagging language or locale constraints when the skill does not offer a language choice or clearly justify the restriction.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The file implements a broad, reusable AI client for chat, image, video, upload, and task polling rather than a narrowly scoped try-on-only skill. In agent environments, this scope mismatch increases the attack surface and can enable unintended capabilities to be invoked outside the declared apparel-preview purpose, weakening policy boundaries and review assumptions.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The module advertises itself as a generic AI capability tool, contradicting the specialized try-on skill metadata. This kind of deceptive or overly broad packaging makes it easier to smuggle unrelated capabilities through review and causes downstream systems to trust a narrower purpose than the code actually provides.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

This code file contains natural-language descriptions, help text, and operational instructions entirely in Chinese, starting in the module docstring and continuing throughout the CLI. Under the policy rule, forcing a specific language without offering the user a locale choice is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

Exposing text-chat capability is inconsistent with the declared virtual try-on image preview purpose and creates a hidden general-purpose model interface. If an agent or wrapper can reach this path, users may leverage the skill for arbitrary prompt processing or policy evasion under the guise of a specialized image tool.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

Video generation materially expands capability beyond the stated try-on preview image workflow. This broadening can be abused to create unauthorized or misleading synthetic media and defeats the expectation that the skill is limited to static apparel preview generation.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

Querying user account information and wallet balance is unrelated to generating try-on previews and introduces unnecessary access to potentially sensitive account data. Even if limited, this violates least privilege and expands what a compromised or misused skill can reveal.

Content

No source excerpt is available for this finding.