Security audit

AI KEY·戳不戳得中人

Security checks for vulnerabilities and agentic risk

Overview

This writing-analysis skill fits its stated purpose and openly saves task-related feedback locally, with no evidence of hidden harmful behavior.

Expect the skill to retain corrected judgments and performance figures in your vault. Specify any confidential material that must not be saved and review the designated memory folder. The shared AI KEY conventions are not bundled; inspect them separately before use, since this assessment covers only the supplied files.

A.I.G

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

SkillSpector

Vulnerability Patterns

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
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (2)

The skill instructs the agent to persist user feedback and performance metrics without warning the user.

Medium

Category: Output Handling
Confidence: 88% confidence
Finding: The closing instruction directs the agent to persist user corrections and numerical performance data without an explicit consent check or data-minimization requirement in the supplied file. These records could retain unpublished editorial judgments or commercially sensitive analytics beyond the current conversation and expose them through later memory use or shared storage. The task-related local destination limits the concern—there is no demonstrated external exfiltration—and safeguards in the referenced but unavailable conventions could mitigate it.
Content: No source excerpt is available for this finding.

Mode C directs the skill to supply openings despite the manifest explicitly excluding opening-writing.

Low

Category: Excessive Agency
Confidence: 96% confidence
Finding: The manifest at L004 and L006 excludes writing openings and assigns that work to aikey-hook. However, L194 permits discussing openings after user confirmation, and L206 explicitly instructs adding suitable openings when creation continues, without handing off to that skill.
Content: No source excerpt is available for this finding.