Back to skill

Security audit

学习系统协调器

Security checks for vulnerabilities and agentic risk

Overview

ClawScan’s verdict is benign with medium confidence: the scans found one scope mismatch involving consent based profile writeback, while the other assessed areas were marked okay; coverage was partial because two local references were unresolved.

The verdict is benign, but consider whether you are comfortable with the reported profile writeback scope mismatch before installing. Confidence is medium, and evidence is incomplete: Skillspector marked analysis partial because two references in SKILL.md did not resolve to bundled artifacts.

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 (1)

The operational instructions permit consent-based profile writeback despite the manifest's explicit no-writeback boundary.

Medium
Category
Excessive Agency
Confidence
95% confidence
Finding

The manifest at L003 limits the coordinator to routing, deduplication, and summarization and explicitly excludes cross-SKILL writeback. However, L436 instructs it to write necessary summaries back to the learning-DNA profile when the user agrees; consent does not make this operation consistent with the declared scope. This also conflicts with L445, which states that the coordinator does not issue writebacks or store profiles.

Content

No source excerpt is available for this finding.