Back to skill

Security audit

30天学习计划制定师

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious because scanners report privacy and scope conflicts around accessing and sharing learning-profile data, though they report no executable files or installation concerns.

Review the skill's profile-data consent and sharing rules before installing, especially whether you are comfortable with broader learning metrics being requested and the schema allowing profile writeback. The scan covered all listed components, but these reported scope and privacy conflicts remain unresolved.

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
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • 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 (4)

The profile retrieval workflow requests broader learning data than the manifest's explicitly limited summary.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

The retrieval workflow includes all-subject error-frequency distributions and recent learning activity, exceeding the explicitly limited summary described in the metadata and consent prompt. Although profile access requires consent, that consent does not clearly disclose these additional metrics, creating a data-minimization and informed-consent gap involving minors' learning records.

Content

No source excerpt is available for this finding.

The parent-dashboard consent check conditionally permits emotional judgments that the documentation categorically prohibits.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

The dashboard checklist conditionally permits emotional judgments when emotion-sharing consent is enabled, whereas other sections categorically exclude them from this skill's parent-facing output. This conflicting generation rule could cause the agent to disclose emotional interpretations outside the promised task-and-completion scope, although the independent consent checks reduce the risk of wholly unauthorized disclosure.

Content

No source excerpt is available for this finding.

Template prerequisites mandate learning-profile use despite the manifest making access opt-in.

Medium
Category
Excessive Agency
Confidence
93% confidence
Finding

The template instructions require placeholders to be filled with profile data and mandate historical completion times without conditioning those requirements on profile-access consent. This creates a competing workflow that could induce profile retrieval even when the student declined it, undermining the skill's default-off privacy controls.

Content

No source excerpt is available for this finding.

The handover contract exposes learning-profile writeback despite the manifest limiting profile access to reading selected summaries.

Medium
Category
Excessive Agency
Confidence
90% confidence
Finding

The manifest describes consent-gated reading of a narrowly defined learning-profile summary, not modification of that profile. L027–L032 enable subject_profile_writeback, and L054–L072 define a patch to extensions.plans, exposing a profile-mutation handover beyond that stated read-only scope. This schema establishes the writeback capability, although it does not demonstrate that a write is executed at runtime.

Content

No source excerpt is available for this finding.