ClawHub
Back to skill

Security audit

Phy Skill Creator

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed workflow for creating and evaluating skills, with some operational cautions around local scripts, subagents, and a temporary review viewer.

Before installing, expect this skill to create local eval workspaces, run Python helper commands, spawn evaluation agents where available, and optionally launch a temporary local review UI. Verify that any referenced helper scripts come from the intended skill package or trusted environment, and stop the viewer process when review is complete.

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
  • 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 (4)

Vague Triggers

High
Confidence
97% confidence
Finding
The instruction to make skill descriptions intentionally 'pushy' encourages overbroad triggering criteria, which can cause the assistant to invoke this skill for adjacent or unrelated user requests. That can degrade routing integrity, override more appropriate skills, and create a confused-deputy-style safety problem where expansive instructions get inserted into contexts that did not warrant them.

Session Persistence

Medium
Category
Rogue Agent
Content
4. **Launch the viewer** with both qualitative outputs and quantitative data:
   ```bash
   nohup python <skill-creator-path>/eval-viewer/generate_review.py \
     <workspace>/iteration-N \
     --skill-name "my-skill" \
     --benchmark <workspace>/iteration-N/benchmark.json \
Confidence
88% confidence
Finding
Using `nohup` to detach the review generator creates session-persistent background execution that can outlive the active task and user awareness. Even though the apparent goal is convenience for long-running review workflows, persistence increases operational risk by leaving orphaned services/processes running and makes side effects harder to audit or control.

Context-Inappropriate Capability

Low
Confidence
79% confidence
Finding
The manifest describes authoring, modifying, evaluating, and optimizing skills, but does not mention controlling the local browser or starting review servers. While review UX can support evaluation, directly opening browser content and managing a server process is an extra capability beyond the core stated purpose.

Context-Inappropriate Capability

Low
Confidence
76% confidence
Finding
The instructions explicitly start a viewer with `nohup ... &`, capture a PID, and later kill the process, and also run an optimization loop in the background. Those are operational process-control capabilities that are not described in the manifest, which only frames the skill in terms of creating, improving, measuring, and optimizing skills.

Static analysis

No suspicious patterns detected.