Back to skill

Security audit

得到大脑(原 Get 笔记)

Security checks for vulnerabilities and agentic risk

Overview

The connection workflow also synchronizes five other AI skills without separate approval, contradicting the skill’s stated limits.

Before connecting, require separate approval for synchronizing other skills and a clear list of affected files and overwrite behavior. Browser authorization enables access to your GetNote account. All bundled documentation was inspected, but the external CLI’s actual behavior remains unverified.

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 (2)

The connection workflow synchronizes other Skills despite an explicit prohibition on modifying other Skill files.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The first-connection workflow requires getnote setup, explicitly described as synchronizing five domain Skills, despite the stated restriction against modifying other Skill files. This expands a connection request into persistent changes to agent instructions without separate authorization; protecting only the current standalone Skill does not protect the other Skills. The official-CLI context suggests an integration-design mistake rather than malicious intent, but does not eliminate the authorization gap.

Content

No source excerpt is available for this finding.

The connection workflow installs or synchronizes agent Skills despite the manifest's promise not to install or update Skills autonomously.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

The initial connection procedure automatically invokes getnote setup, and this reference confirms that the command installs or synchronizes domain Skills for local agents. A request to connect or authenticate therefore triggers persistent agent-instruction changes beyond the requested account connection, contrary to the advertised no-autonomous-Skill-update boundary. The explicit separation of the standalone Skill reduces its overwrite risk but leaves the five domain Skills exposed to unapproved changes.

Content

No source excerpt is available for this finding.