Back to skill

Security audit

Deepmine 5.1

Security checks for vulnerabilities and agentic risk

Overview

This is a Chinese-language guided questioning skill with no code execution or external access; its optional tax/compliance risk spotting is disclosed but should not be treated as professional advice.

Install this if you want a Chinese structured questioning assistant for clarifying ideas or requirements. Expect it to ask repeated follow-up questions and append conversation state metadata. For tax, legal, or compliance topics, treat its risk labels as prompts for professional review, not as authoritative advice.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (21)

Vague Triggers

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The trigger conditions are extremely broad and map to common phrases like '帮我做个方案' or '我不知道该怎么办', which can cause the skill to activate for many unrelated user requests. Because this skill imposes a rigid multi-step questioning workflow and internal state handling, over-triggering can hijack normal interactions, derail safer default behaviors, and increase the chance of collecting unnecessary sensitive context from users.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The file implements a domain-specific财税经营 extraction and risk-screening workflow, which materially differs from the manifest’s stated Socratic deep-questioning behavior. This kind of hidden capability can cause the agent to invoke an unexpected compliance-analysis function, expanding data handling and decision influence beyond what users and integrators intended.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill performs specialized tax/compliance risk assessment, including criminal-red-line detection and legal-risk labeling, despite being presented as a generic reflective-questioning skill. This mismatch is dangerous because it enables undisclosed high-stakes analysis that may process sensitive business statements and steer users on legal/compliance matters without appropriate transparency, consent, or qualification controls.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The README presents all user-facing instructions and descriptions exclusively in Chinese, and nowhere indicates that users may choose another language or that the skill is intended only for a Chinese-speaking audience. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The skill metadata and behavior are entirely Chinese-language oriented, with Chinese trigger phrases and no indication that language should follow the user's preference. This can cause the skill to engage in the wrong locale, confusing users, degrading transparency and consent, and making it harder for users to understand what data they are providing during a probing questioning flow.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill title and core instructions require operation entirely in Chinese, including role framing and response schema content, but do not offer the user any language or locale choice. This is a natural-language policy concern because it imposes a specific language behavior without opt-in or a documented regional justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The file is written as mandatory behavioral instructions in Chinese and repeatedly specifies exact response style and wording, but nowhere offers the user a language choice or states that the skill is limited to Chinese-language interactions. This creates a natural-language locale policy concern because the skill appears to enforce a specific language by default rather than through user opt-in.

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 is entirely written as Chinese-only operating instructions, including mandated reply patterns and example outputs, which implies the skill will respond in Chinese by default. The policy allows locale constraints only when users are given a choice or when the restriction is clearly documented and justified; neither appears here.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The handler contains mutually incompatible control instructions: one high-priority section says each reply must include a confirmation plus the next question, while later breakthrough-specific rules require restating and explicitly not asking a follow-up in the same turn. In an agent setting, such conflicts make behavior nondeterministic and can cause the model to violate conversation safety or workflow guarantees, especially when orchestrator state depends on a specific turn structure.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

Later sections reinforce a two-sentence pattern of 'directly ask, or one confirmation + one question' while the earlier execution rules prohibit asking any follow-up after restatement. This contradiction can be exploited indirectly by adversarial or edge-case inputs to steer the model into breaking one of its own policies, reducing reliability and making downstream state transitions inconsistent.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The manifest describes a Socratic deep-questioning skill whose end product is a structured conclusion document derived from the user's own judgments. This handler instead instructs the model to produce a '需求摘要' explicitly '供后续生成思维导图使用', shifting the skill from reflective conclusion capture toward requirements elicitation and artifact preparation for another workflow.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The manifest says the skill is triggered for thinking clearly, reviewing experiences, generating plans, and clarifying needs, while avoiding giving answers and surfacing the user's own judgment. The mandated sections—business background, core features, user groups, technical constraints, time/budget, and acceptance criteria—are specifically a requirements-document structure, which does not fit many listed use cases such as retrospectives or personal dilemmas.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The handler content is entirely Chinese and operationally constrains responses to Chinese without any user language negotiation or documented locale requirement. This can exclude or mislead users who interact in another language, causing incorrect task execution, consent misunderstandings, or inaccessible guidance; in a questioning/decision-support skill, that can materially degrade user outcomes.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

This markdown file contains natural-language operating instructions exclusively in Chinese and does not indicate that the user may choose another language or that the skill is limited to a Chinese-language context. Under the policy rule, forcing a specific language without opt-in is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The instructions are written as mandatory behavior in Chinese, directing the skill to operate as a router and output only JSON, but they do not offer any user language or locale choice. This creates a natural-language policy concern because the skill appears to enforce a specific language context without explicit opt-in or documented justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

This markdown file is entirely written in Chinese and provides no indication that the skill supports other languages or that Chinese-only operation is a documented, justified regional constraint. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The file requires a Chinese-only state/output format with no language choice or opt-in, which can create an unsafe mismatch between user language expectations and system behavior. In a multi-language agent environment, this can cause hidden instruction-channel confusion, reduce user comprehension of structured metadata, and make it harder for downstream reviewers or safety systems to inspect what the model is appending.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

SQP-3 applies to all file types and covers language or locale policy violations. This markdown file presents all user-facing content exclusively in Chinese and does not mention that the language is optional, selectable, or intended only for a Chinese-speaking audience.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

This file is primarily in English, but lines L25-L33 switch to Chinese for substantive licensing terms. Under the policy, forcing a specific language for important instructions or terms without user opt-in or clear justification is a natural-language locale violation.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

L127 says replies must be '不评判' and not express a stance, but the required response examples at L31-L33 and the mandated structure at L27 encourage evaluative affirmations such as '这个想法不错' and '方向很清楚'. This is an internal documentation contradiction about the intended interaction behavior.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
92% confidence
Finding

The title and all guidance in this file are written exclusively in Chinese, and there is no indication that users may choose another language or that the skill is intentionally limited to a Chinese-language context. Under the language/locale policy, forcing a specific language without opt-in is a natural-language policy concern.

Content

No source excerpt is available for this finding.