Back to skill

Security audit

Pua

Security checks for vulnerabilities and agentic risk

Overview

This skill is disclosed as a pressure-style productivity mode, but it can activate on broad phrases and persist behavior rules across future sessions.

Install only if you intentionally want a forceful, Chinese corporate-style work mode and are comfortable with automatic activation. Before using it, remove or disable the auto-trigger list and the ~/.pua/evolution.md persistence mechanism, and require explicit user approval before broad searches, repository scans, endpoint calls, or extra work outside the requested task.

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
Findings (2)

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:25
Finding

Automatic Activation Hijacks Agent Goals and Expands Task Scope

Content
View full analysis
Remediation
View remediation

T02 · Agent Memory Poisoning

Error
Location
SKILL.md:321
Finding

Persistent Behavioral Rules Poison Future Agent Sessions

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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 (4)

Vague Triggers

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The metadata description defines automatic activation on broad signals like frustration, passive behavior, quality complaints, and common phrases such as 'try harder' or '加油'. These triggers are highly ambiguous and likely to fire during ordinary conversation, causing the skill to override the normal assistant behavior without clear consent. In context, that is especially risky because this skill materially changes tone, pressure, and decision-making expectations across all task types.

Content

No source excerpt is available for this finding.

Vague Triggers

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The explicit auto-trigger list contains common Chinese and English phrases that users may say casually, sarcastically, or in unrelated contexts. Because there are no exclusion rules or confirmation steps, the skill can activate unexpectedly and impose coercive behavior patterns the user did not knowingly request. The risk is heightened by the skill’s broad applicability to 'All task types' and its pressure-escalation mechanics.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The skill is predominantly written to enforce a Chinese-language, Chinese-corporate style and repeatedly instructs output patterns and phrasing without requiring user opt-in. For users who did not request Chinese output or this cultural framing, the skill can degrade usability, misalign with user intent, and interfere with accessibility or policy-compliant communication. The danger is amplified because the skill is designed to auto-trigger, meaning the language/style shift may happen unexpectedly.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill directs the agent to store cross-session performance and behavioral data in ~/.pua/evolution.md, which is unrelated to the core function of applying a response style. Persistent state can create unauthorized memory, privacy leakage, and hidden behavior shaping across unrelated future tasks. In context, this is more dangerous because the stored data is specifically used to intensify pressure and alter future conduct without fresh user consent.

Content

No source excerpt is available for this finding.