Back to skill

Security audit

Skill Collision Engine

Security checks for vulnerabilities and agentic risk

Overview

This skill is a review case because it describes broad automatic skill generation, self-evolution, external knowledge ingestion, and use of conversation records without clear consent, scoping, or rollback controls.

Install only if you are comfortable with a meta-skill that may read skill logs, combine external materials, and automatically produce or update skill-related artifacts. Do not point it at private conversations, sensitive workspaces, or regulated data unless you add your own consent, redaction, source allowlist, human review, and rollback controls.

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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (8)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

Listing human conversation records as a preferred knowledge source without any privacy, consent, retention, or sensitivity guidance is a substantive security and privacy flaw. Conversation logs often contain personal data, credentials, business secrets, or regulated information, and this engine’s synthesis behavior could propagate that data into derived skills or artifacts.

Content

No source excerpt is available for this finding.

Self-Modification

High
Category
Rogue Agent
Confidence
94% confidence
Finding

The documented self_evolve capability explicitly changes future behavior based on prior logs, constituting self-modification without described safeguards, review controls, or rollback procedures. In a skill factory context, this can compound errors, amplify poisoned feedback, or drift into unsafe behavior across subsequent runs and generated skills.

Content

Scanner excerpt · SKILL.md (reported line 162)May include surrounding context.

)

4. 工厂元进化

engine.self_evolve() # 读取验证日志,优化碰撞参数

text

## 当前 Skill 碰撞矩阵(已发布6个 Skill)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The primary descriptive content is presented in Chinese ('Skill碰撞引擎 —— 让无穷层级的认知框架自动碰撞产生新Skill的元引擎') and the rest of the README continues in Chinese, but there is no indication that the skill is China-specific or that users can opt into another language. This creates a natural-language policy concern because it imposes a locale/language requirement without user choice or documented justification.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill frames itself as able to collide arbitrary skills and external knowledge sources with few stated boundaries, which creates unsafe trigger ambiguity and encourages use in situations involving sensitive or untrusted inputs. In a meta-skill that can aggregate other skills and data sources, broad activation scope increases the chance of unintended access, overreach, or unsafe autonomous behavior.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The documentation does not clearly distinguish when this engine should be used instead of ordinary composition, leaving invocation semantics underspecified. For a system advertised as producing emergent new skills, lack of bounded activation criteria can cause operators or agents to invoke it too broadly, amplifying downstream risk from unsafe synthesis or data mixing.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill describes reading verification logs and colliding with external sources, but gives no warning or controls around local file access, network retrieval, or ingestion of untrusted content. In this context, the engine is a meta-framework that may combine internal workspace data with external materials, making accidental data exposure, prompt injection via content, or unauthorized access materially more dangerous.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The verification protocol content is written entirely in Chinese, including headings, checklist items, and the conclusion template, with no indication that users may choose another language. The stated policy flags language or locale constraints when a skill forces a specific language without user opt-in or clear justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
81% confidence
Finding

Most of the natural-language instructions and descriptions are in Chinese, while the file title and some identifiers are in English. Because the skill does not state that it is Chinese-only or offer a language/locale choice, it may impose a language preference without explicit user opt-in.

Content

No source excerpt is available for this finding.