Back to skill

Security audit

数学解题教练

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this math tutoring skill benign with medium confidence: the scan reports no AIG findings, while Skillspector flags some scope mismatches and a language-choice issue; one referenced item was missing, so the evidence is incomplete.

The verdict is benign, but the confidence is medium and the scan is partial: a local reference in SKILL.md could not be matched to a bundled artifact. Consider whether the reported scope mismatches and language-choice issue matter for your use, and treat the verdict as an assessment of the available evidence rather than a guarantee of safety.

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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (7)

The systemic-error branch directs cross-problem root-cause analysis despite explicitly stating that this skill does not perform it (L043–L062).

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

L043 says multi-problem attribution must be delegated to the correction-notebook and math-error-DNA skills, with this skill only using their conclusions. However, L048–L062 instruct the assistant to solicit three problems, compare them, identify their shared sticking point, and summarize the root cause itself, directly contradicting that delegation boundary.

Content

No source excerpt is available for this finding.

Ae4

Medium
Category
analysis-evasion
Confidence
80% confidence
Finding

Suspicious Unicode normalization or mixed-script content

Content

No source excerpt is available for this finding.

Ae4

Medium
Category
analysis-evasion
Confidence
80% confidence
Finding

Suspicious Unicode normalization or mixed-script content

Content

No source excerpt is available for this finding.

Ae4

Medium
Category
analysis-evasion
Confidence
80% confidence
Finding

Suspicious Unicode normalization or mixed-script content

Content

No source excerpt is available for this finding.

A tutoring prompt requires the student to explain the mathematical relationship in Chinese without offering a language choice.

Low
Category
Output Handling
Confidence
98% confidence
Finding

The prescribed prompt explicitly says “用中文说” (“say it in Chinese”). This imposes a specific natural language without user opt-in or a documented justification; it is not merely an instruction to explain clearly.

Content

No source excerpt is available for this finding.

The workflow restricts generated practice problems to middle-school material despite the manifest explicitly including high-school mathematics.

Low
Category
Excessive Agency
Confidence
98% confidence
Finding

The manifest covers both middle school and high school, including compulsory and selectively compulsory high-school content. L182 instead requires generated problems to remain within the middle-school curriculum, preventing this core workflow from providing appropriately matched practice for high-school problems.

Content

No source excerpt is available for this finding.

The workflow assigns mathematics error subtypes even though the manifest delegates error-subtype analysis to another skill.

Low
Category
Excessive Agency
Confidence
95% confidence
Finding

The manifest explicitly excludes error-subtype analysis and routes it to xiaozhi-math-error-dna. L266 nevertheless directs this workflow to populate subtypeId using mathematics error-subtype codes, implementing that excluded classification within the coaching session; session-only storage does not resolve the functional scope mismatch.

Content

No source excerpt is available for this finding.