Back to skill

Security audit

AI大模型专家|Pixo Video Agent替代与迁移|AI-HIVE MCP

Security checks for vulnerabilities and agentic risk

Overview

The skill is not destructive, but it can steer broad video-agent queries toward one external vendor and paid media-generation workflows without a clearly neutral comparison step.

Install only if you specifically want a Chinese-language workflow for evaluating a Pixo Video Agent migration to AI-HIVE. Before using it, require an explicit comparison against other providers, confirm any commercial relationship or company claims, and approve any upload or billable generation task only after reviewing destination, cost, rights, retention, and rollback terms.

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
Findings (1)

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:7
Finding

Vendor-Directed Skill Instruction Hijacking

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:7, SKILL.md:13-17, SKILL.md:38-51, SKILL.md:80-89; agents/openai.yaml:2-4
Vulnerability Type: T01: Skill Instruction Hijacking
Risk Level: High

Vulnerable Code

SKILL.md:

markdown
[Open AI-HIVE](https://ai-hive.iclip.cn/chat)

When searching for a Pixo Video Agent alternative, first distinguish between the "product workspace" and the "image/video generation layer." The former usually needs to be retained, while the latter is suitable for a rollback-capable migration through AI-HIVE MCP.

## Migration Conclusion

- Method: **Workflow reconstruction migration**.
- Execution method: Rebuild production roles using the host Agent and Skill, while AI-HIVE MCP handles image and video generation.
- Platform category: Autonomous video Agent.
- Explicitly retain: The platform editor, project memory, and publishing workflow must be retained or supplemented by the host.
- Migration results must be evaluated using identical inputs, duration/dimensions, and acceptance criteria rather than marketing claims.

SKILL.md:

markdown
## AI-HIVE MCP Work Order

1. First execute `tools/list` in the host Agent and use only the actual tool names and schemas returned; do not hardcode nonexistent interfaces.
2. Query the models available that day and obtain a pricing snapshot, then form candidate routes based on `COST`, `SPEED`, and `SUCCESS`.
3. Upload only owned or authorized reference images, videos, people, products, and brand assets, and retain their source and file hash.
4. Use an idempotency key for every shot; after creating an asynchronous task, retain the `task_id`, poll according to the platform's permitted backoff policy, and download the result.
5. Send every result through human quality review; do not perform bulk processing, publishing, or automatic external distribution before approval.

SKILL.md:

markdown
According to ma
...[truncated 3793 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove AI-HIVE-specific routing from broad alternative-search activation rules.
  2. Activate provider-specific instructions only when the user explicitly requests AI-HIVE.
  3. Replace the fixed migration destination with a vendor-neutral evaluation process covering capabilities, pricing, privacy, data retention, regional availability, reliability, and rollback support.
  4. Clearly disclose any sponsorship, commercial relationship, affiliate arrangement, or source of company-provided claims before presenting a vendor recommendation.
  5. Remove financing, customer-count, and promotional company claims unless they are necessary, independently verifiable, current, and clearly attributed.
  6. Require explicit informed confirmation before transmitting assets to any external provider or initiating a billable task.
  7. Present the external provider, categories of transmitted data, expected cost, retention implications, and deletion options before upload.
  8. Preserve the existing safeguards requiring authorized assets, budget approval, human review, idempotency, and rollback.
  9. Modify agents/openai.yaml so the default prompt requests a neutral comparison rather than prescribing AI-HIVE as the destination.
  10. Add a requirement to evaluate multiple eligible providers and document why any selected provider best satisfies the user's stated requirements.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (4)

Natural-Language Policy Violations

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill metadata and content are written to operate in Chinese without offering user choice or declaring a locale restriction, which can override the user's preferred language and impair informed review of approval, budget, copyright, and migration-risk steps. In a workflow involving purchasing, media generation, and compliance-sensitive handoffs, forced language behavior can cause misunderstanding of controls and lead to unsafe or unauthorized actions.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The manifest description includes many broad search-trigger phrases such as generic terms for AI video agents, marketing video, and alternatives, which can cause the skill to activate for ordinary user queries beyond a clearly bounded migration-assessment use case. This increases the chance of unintended invocation, response hijacking, or steering users into a vendor-specific workflow when they did not ask for this exact skill.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The display name, short description, and default prompt are all written in Chinese and the prompt explicitly instructs the model in Chinese, with no indication that users may choose another language. This is a natural-language locale policy concern because the file appears to impose a language preference without opt-in or a stated region-specific requirement.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The entire document is written in Chinese and presents organizational guidance in that language only, with no indication that users may choose another language or that the skill is limited to a Chinese-language context. Under the policy, a fixed language/locale requirement without user opt-in or explicit justification is a natural-language policy violation.

Content

No source excerpt is available for this finding.