Back to skill

Security audit

MiniMax H3 Studio

Security checks for vulnerabilities and agentic risk

Overview

This is a document-only Chinese MiniMax H3 creative workflow skill that is coherent with its stated purpose and does not show hidden credential access, persistence, destructive behavior, or deceptive execution.

Install this if you want a Chinese-oriented MiniMax H3 prompt and video-production workflow. Before using full media generation, confirm the selected provider, costs, rights to uploaded brand/product assets, and whether you want Chinese or another output language; the skill's defaults are opinionated but disclosed.

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
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • System Prompt LeakageDirect Leakage, Indirect Extraction, Tool-Based Exfiltration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (56)

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · .gitignore (reported line 4)May include surrounding context.

text
.DS_Store
__pycache__/
*.pyc
.env
.env.*
exports/

Vague Triggers

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

The confirmation gate is weakened by treating vague conversational phrases like 'continue', 'next', '继续', or '下一步' as permission to generate. In a multi-step agent flow, users often use these phrases to request more planning or clarification, so the skill may begin media generation, web use, or asset processing without clear informed consent to proceed past the hard confirmation checkpoint.

Content

No source excerpt is available for this finding.

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · references/h3-prompt-writing/官方原文/SKILL.md.txt (reported line 31)May include surrounding context.

text
Read `references/ref-en.txt` for label rules, retention analysis, and complete examples.

## Output Rules

- Write rewrite sections in English; preserve dialogue, lyrics, and visible scene text in their original language.
- Describe each shot by composition, subjects, environment, actions, camera, sound, and the exact point where referenced content appears.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The usage example instructs the skill to generate content with '全部使用中文', reinforcing a mandatory language restriction. This is a natural-language policy concern because it normalizes a fixed language requirement rather than offering the user a choice.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The README states that prompt text, titles, and shot descriptions are uniformly in Chinese, which imposes a language constraint as a default requirement. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation unless the locale constraint is clearly documented and justified as region-specific.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The instruction says all prompt text, structural titles, character settings, shot descriptions, and interactions must use Chinese. This is a language-policy constraint presented as a default requirement, and it does not offer users a general choice of language except for limited cases like preserving already-locked foreign-language text.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The text states support for Chinese prompts and instructs the skill to output fully Chinese-structured prompts, which imposes a language choice by default. Under the policy, language constraints should offer user choice unless clearly justified as region-specific.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

This manifest file describes invocation behavior in a way that leaves trigger scope and routing ambiguous. Because it references selecting a workflow from nine capabilities based only on the user's needs, without explicit constraints or examples, the skill may be invoked or routed too broadly.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The document states that unless the user explicitly specifies otherwise, all outputs must use a fixed '皮克斯感 3D 卡通渲染' style. This imposes a specific stylistic/locale-like preference by default rather than presenting it as a user choice, which matches the policy concern about forcing a setting without opt-in.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The instruction hard-codes the English phrase “continue” as a control signal, which creates a language-specific interaction policy. This can disadvantage users operating in other languages and does not provide any opt-in or alternative locale handling.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The guideline hardcodes reference markers using Chinese placeholders like 角色名 and 场景名, which indicates a required locale-specific naming convention. Because the document does not offer user opt-in or an alternative language choice, this is a natural-language locale policy issue.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The copy-paste skeleton provides the canonical example in Chinese (奶奶把苹果筐递给 Mia, 奶奶弯下腰从中岛拿起苹果筐, Mia 接住筐转身,门铃响起) and does not state that language is optional. In a guideline file, such examples can function as implicit instruction to produce output in that language, which violates the language-choice policy when no opt-in is provided.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The mandatory image-label specification again requires placeholders 角色名 and 场景名 for storyboard labels, reinforcing a fixed Chinese locale. Since the file gives no opt-in or justification for that constraint, it presents a language/locale policy concern.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly instructs the agent to read and follow an external '统一中文规范' before execution, which creates a prompt-injection and hidden-instruction surface outside the current file. Because referenced content may contain additional constraints or conflicting rules, the agent can be steered by untrusted external text before handling the user request, reducing transparency and making security review incomplete unless the dependency is also analyzed.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

L040 规定当品牌素材或来源文案主要为英文时,默认旁白和画面文案使用英文,即使用户以中文交流也如此。这属于语言/locale 的默认强制策略,缺少明确的用户 opt-in 或语言选择,符合自然语言政策中的语言选择违规情形。

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The skill says to use it when a user 'has a logo, product images, screenshots, a website link, or just a clear idea and wants the agent to turn those materials into a clean brand reel.' The phrase 'or just a clear idea' makes the activation scope quite broad and lacks negative examples or tighter trigger constraints, which could cause unintended invocation for generic video-creation requests.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The language rule says to default narration and on-screen copy to English when brand assets are primarily English unless the user explicitly asks for Chinese localization. This imposes a specific output language by default rather than offering the user a language choice, which is a natural-language locale policy concern.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The language rule states that if brand materials are mainly English, the narration and on-screen copy should default to English unless the user explicitly requests Chinese localization. This imposes a specific output language without user opt-in, even when the user is interacting in Chinese, which matches the policy category for language/locale violations.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

Lines L025-L027 switch the prompt template into Chinese imperative instructions for the global style baseline. This imposes a specific language/locale in the skill content without any opt-in, alternative language path, or justification that the skill is intended only for Chinese-language use.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The file is entirely written as a prescriptive Chinese workflow and does not indicate that users may choose another language or locale. Under the policy, forcing a specific language without opt-in is a natural-language policy violation unless a justified regional constraint is documented.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The prompt hard-codes a default exclusion of female characters unless explicitly requested, which introduces discriminatory behavior into the system's output policy. In a content-generation skill, this can systematically bias outputs, restrict representation, and create harmful or non-compliant behavior when used in production workflows.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

Line L01 states that the prompt body and structure must be in Chinese and that a unified Chinese specification must be read before execution. This is a natural-language locale policy constraint, and the file does not present it as a user choice or opt-in language preference.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

This plain-text file contains natural-language instructions requiring overall_soundscape and non_diegetic_music to be written in English. That forces a specific language choice rather than offering the user a language/locale option or documenting a justified region-specific constraint.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

Line L005 instructs writers to produce all six rewrite sections in English and only preserve the original language for limited content like dialogue, lyrics, and visible text. This imposes a fixed language requirement without offering a user choice or documenting a justified region-specific constraint.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The instruction explicitly requires rewrite sections to be written in English, which imposes a language constraint regardless of the user's preferred language. This is a natural-language policy concern because the file does not offer an opt-in, alternative language choice, or a documented region-specific justification.

Content

No source excerpt is available for this finding.