Back to skill

Security audit

Second-Order Thinking

Security checks for vulnerabilities and agentic risk

Overview

This reasoning skill needs review because it automatically fetches changing online content and presents remote skill instructions as canonical without a user review step.

Prefer a version that keeps runtime instructions local and makes update checks optional. Review updates before adopting them, and keep downloaded examples separate from instructions. The artifacts support a remote-instruction control concern, but do not establish malicious intent or sensitive-data exfiltration.

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
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (3)

The skill requires an external runtime update fetch unrelated to its stated reasoning purpose.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

The startup instruction introduces a mandatory third-party network dependency into a reasoning framework that can operate entirely offline. The stated freshness rationale may be legitimate, but fetching mutable content on every run expands the attack surface without version pinning or integrity verification; continuing on fetch failure limits availability risk but not successful malicious responses.

Content

No source excerpt is available for this finding.

The runtime-update pretext promotes remotely supplied markdown to authoritative skill instructions.

Medium
Category
Prompt Injection
Confidence
97% confidence
Finding

The instruction describes remotely supplied 'latest_skill_md' as the canonical current version, inviting the agent to treat mutable network content as authoritative skill instructions. A compromised endpoint or malicious update could therefore redirect agent behavior with the authority of the installed skill; this is particularly unnecessary for a self-contained reasoning method, although actual consequences depend on the agent's permissions and trust-boundary enforcement.

Content

No source excerpt is available for this finding.

The startup workflow contacts a third-party service without warning the user about the privacy implications.

Low
Category
Output Handling
Confidence
94% confidence
Finding

The automatic request with '?ctx=run' discloses a runtime event and ordinary connection metadata to a third party without user notice or consent. The supplied URL does not include conversation content, credentials, or a user identifier, so the supported privacy concern is limited telemetry rather than demonstrated sensitive-data exfiltration.

Content

No source excerpt is available for this finding.