Back to skill

Security audit

Cjl Autoresearch Cc

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed iterative editor for improving user-selected text, skills, prompts, or workflows, with some broad activation language but no hidden execution or persistence.

Install only if you want an agent to help edit selected prompts, documents, skills, or plugins. Confirm the target carefully, especially for ~/.claude skill or plugin paths, and review the batch summary before allowing additional rounds.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
  • 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
Findings (10)

Vague Triggers

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

The semantic trigger examples are broad natural-language phrases like asking to optimize a skill, prompt, article, or system, which can overlap with ordinary user requests and cause accidental invocation. In an agent-skill environment, overly permissive triggers can make the skill activate in contexts the user did not explicitly intend, leading to unexpected modification-oriented behavior on arbitrary content.

Content

No source excerpt is available for this finding.

Vague Triggers

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The semantic triggers are broad enough to match many ordinary requests for improvement, refinement, or polishing, which can cause this skill to activate when the user did not specifically intend to invoke it. In an agent setting, overbroad activation increases the chance of prompt-routing mistakes, unexpected behavior, and interference with other skills or baseline assistant behavior.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · README.md (reported line 39)May include surrounding context.

md
### The Solution: Mutation Testing

Instead of asking AI to rewrite everything (which often makes things worse), we:

1. **Make one small change** at a time
2. **Test each change** against realistic scenarios

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The mode-detection rules map loosely phrased inputs and paths to high-level modes without clear constraints, precedence rules, or counterexamples. This ambiguity can cause the agent to misclassify arbitrary input as a Skill, Plugin, Prompt, Workflow, or System target, which is risky because each mode may imply different levels of access, mutation, or persistence.

Content

No source excerpt is available for this finding.

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · README.md (reported line 189)May include surrounding context.

最有效变异: Type D (收紧语言), Type A (添加约束)


最终内容保存至: ~/.claude/skills/coding-standards/SKILL.md

text

---

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · README_zh.md (reported line 189)May include surrounding context.

最有效变异: Type D (收紧语言), Type A (添加约束)


最终内容保存至: ~/.claude/skills/coding-standards/SKILL.md

text

---

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The semantic trigger examples are very broad everyday phrases such as asking to 'optimize this skill' or 'make the article smoother', which can cause the skill to activate without an explicit, unambiguous invocation. In an agent environment, overly permissive trigger logic can redirect normal user conversations into autonomous optimization behavior, increasing the chance of unintended file edits or workflow changes.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The keyword triggers include generic phrases such as optimize, improve, refine, and enhance, which are commonly used in normal conversation and are not distinctive to this skill. That makes accidental invocation more likely, potentially causing the agent to enter an optimization loop or impose this skill's workflow on unrelated user requests.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

Defaulting ambiguous requests to Prompt mode creates unsafe fallback behavior because uncertainty should reduce activation confidence, not force a mode selection. This can lead the system to treat unrelated or underspecified inputs as optimization targets, producing unintended handling and making routing behavior unpredictable.

Content

No source excerpt is available for this finding.

Unrestricted Tool Access

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill grants unrestricted tool access without appropriate constraints. An agent with unfettered tool access can perform arbitrary actions including file modification, network requests, and code execution.

Content

Scanner excerpt · references/mutation_strategies.md (reported line 88)May include surrounding context.

md
Balance constraints vs. flexibility.

**Too constrained:** "Use exactly these 3 tools in this exact order"
**Too loose:** "Use any tools you think are appropriate"
**Balanced:** "Use the appropriate tools for the task. Recommended order: tool A, then B, then C unless D is detected"

## Scoring Matrix