ClawHub
Back to skill

Security audit

Meta Research

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed research-workflow assistant that can search, write project files, and run analysis commands, with no evidence of hidden persistence, credential use, exfiltration, or destructive behavior.

Install this when you want an agent to help manage a research project, search the web, write project files, and potentially run analysis code. Prefer project-scoped installation unless you want it available everywhere, and review before allowing expensive compute, public releases, or changes to important research artifacts.

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
  • 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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (3)

Vague Triggers

Medium
Confidence
96% confidence
Finding
The skill declares very broad trigger terms such as 'research', 'analysis', and 'brainstorm', which are common in ordinary user requests and can cause this high-autonomy skill to activate in situations where a narrower skill would be safer or more appropriate. Because the skill is user-invocable and has powerful capabilities including Bash, WebFetch, Write, and Task operations, over-broad matching increases the chance of unintended activation and unnecessary tool use.

Session Persistence

Medium
Category
Rogue Agent
Content
2. **Constrain degrees of freedom early**: lock primary metric, dataset, baseline before
   large-scale runs
3. **Reward null results**: negative findings are logged as valid milestones, not failures
4. **Pre-commit before scaling**: write down the analysis plan before running big experiments
5. **Multiple comparisons awareness**: if testing N models × M datasets × K metrics,
   acknowledge the multiplicity and use corrections or frame as exploratory
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.

Session Persistence

Medium
Category
Rogue Agent
Content
2. **Constrain degrees of freedom early**: lock primary metric, dataset, baseline before
   large-scale runs
3. **Reward null results**: negative findings are logged as valid milestones, not failures
4. **Pre-commit before scaling**: write down the analysis plan before running big experiments
5. **Multiple comparisons awareness**: if testing N models x M datasets x K metrics,
   acknowledge the multiplicity and use corrections or frame as exploratory
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.

Static analysis

No suspicious patterns detected.