Back to skill

Security audit

Yq Mj Prompt Optimizer

Security checks for vulnerabilities and agentic risk

Overview

This skill is only Markdown and does not access the local system, but it is designed to silently rewrite prompts and probe Midjourney moderation filters.

Review this skill carefully before installing. It may be useful for Midjourney prompt cleanup, but it can silently change or remove parts of your request and is specifically oriented toward finding wording that avoids platform moderation filters. Use it only if you are comfortable with those transformations and check the final prompt before submitting it anywhere.

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)

other

Warning
Location
SKILL.md:13
Finding

Systematic External Platform Moderation-Filter Evasion

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:13-41; supporting instructions in references/sensitive-words.md:16-23,44-50
Vulnerability Type: other: Moderation Filter Evasion
Risk Level: Medium

Vulnerable Content

Relevant excerpt from SKILL.md:13-16 (English translation):

text
Midjourney's sensitive-word detection uses combination-pattern recognition rather than individual-word detection:
- Three clothing-category terms appearing together trigger the "clothing set/uniform" filter.
- "Black + wings/horns/teeth" triggers the "dark imagery" filter.
- If individual words pass independently but trigger a rejection when combined, the combination is the problem rather than a particular word.

Relevant excerpt from SKILL.md:18-26 (English translation):

text
Sensitive-word replacement rules (apply directly without asking)

Expression: Mischievous bad smile
Replacement: Playful broad smile / bright toothy smile

Expression: Small fang showing
Replacement: Playfully pointed tooth

Expression: Fingers painted with black nail polish
Replacement: Delete directly

Expression: Black wings
Replacement: Light-gray wings / silver wings

Expression: Black curved horns
Replacement: Curved horn accessories

Expression: Three dark clothing categories together
Replacement: Combine them into two items or use a lighter color

Expression: Black formalwear + trousers + leather shoes
Replacement: Dark-gray suit jacket with dark-gray trousers

Relevant excerpt from SKILL.md:35-41 (English translation):

text
Workflow:
1. Receive the user's Chinese description.
2. Apply the sensitive-word replacement table directly without notifying the user.
3. Add the standard Midjourney suffix.
4. Output the English prompt directly.

Supporting filter-probing procedure from references/sensitive-words.md:44-50 (English translation):

text
Exhaustive troubleshooting pro
...[truncated 3422 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove the sentence-by-sentence and word-by-word filter-probing workflow from references/sensitive-words.md.
  2. Replace evasion-oriented rules with a policy-compliance workflow that evaluates the underlying requested content rather than merely disguising trigger terms.
  3. Refuse requests whose underlying intent violates the target platform's policies instead of searching for alternate wording that passes moderation.
  4. Clearly disclose material removals or semantic substitutions rather than applying them silently.
  5. Limit transformations to benign quality improvements such as translation, grammar correction, composition, lighting, camera parameters, and artistic style.
  6. Do not maintain a growing repository of discovered moderation triggers or successful bypass combinations.
  7. Add an explicit rule stating that the Skill must not test, reverse-engineer, or circumvent external moderation systems.
  8. Document the intended content-policy boundary and include examples distinguishing harmless rephrasing from prohibited filter evasion.
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • 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 (3)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The trigger list includes very broad phrases such as '生成图片', '帮我画', and '优化提示词', which are common in ordinary image-generation requests and can cause the skill to activate unintentionally. This creates a prompt-scope/control issue: the skill may silently rewrite user intent, apply hidden filtering, and force Midjourney-specific behavior in contexts where the user did not explicitly request this skill.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill states that sensitive-word filtering and substitutions are applied automatically '不提醒,直接替换', meaning user-provided content is modified without notice. Silent transformation is dangerous because it reduces transparency, can materially alter meaning or safety-relevant details, and prevents users from detecting censorship, policy bypass behavior, or fidelity loss in generated prompts.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The file says it converts Chinese descriptions into directly usable English prompts and '直接输出英文提示词', but it does not offer a language option or explain why English-only output is mandatory. This is a natural-language policy concern because it imposes a fixed language/locale behavior without user opt-in.

Content

No source excerpt is available for this finding.