Back to skill

Security audit

发布智能体技能_无限

Security checks for vulnerabilities and agentic risk

Overview

The skill has a coherent ClawHub publishing purpose, but it automates local file deletion, desktop input, clipboard changes, and public MIT-0 publication without enough confirmation.

Review this skill before installing. Use it only when you intentionally want an agent to publish a local skill to ClawHub, and manually verify the folder contents, account, license, and final publish action. Avoid running it on folders with important backup files unless the deletion step is removed or explicitly confirmed.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (3)

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:78
Finding

Silent Destructive Deletion of a Local Backup File

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:113
Finding

Unscoped GUI Keystrokes and Destructive Clipboard Replacement

Content
View full analysis
Remediation
View remediation

other

Warning
Location
SKILL.md:179
Finding

Publication Under MIT-0 Without Explicit License Confirmation

Content
View full analysis
" ``` ```powershell & $NODE "...xb.cjs" run --browser default --headed click "@" ``` The first command selects the checkbox whose accessible name contains `MIT-0`. The second command submits the publication after that checkbox enables the publish button. ### Technical Analysis The workflow mandates selecting MIT-0 and proceeding to publication but does not require a separate confirmation that: - The user owns or is authorized to distribute every uploaded file. - MIT-0 is the intended license. - The upload contains only the files the user expects. - The currently authenticated ClawHub account is the intended publisher. - The user approves the final public publication action. License selection is a consequential legal and distribution decision. Treating it as a routine automation step can cause content to be released under terms the user did not intend or lacks authority to grant. ### Attack Path 1. A user asks the agent to publish a supplied skill folder. 2. The workflow uploads the folder. 3. The agent automatically selects the MIT-0 checkbox. 4. The publish button becomes enabled. 5. The agent submits the publication without obtaining explicit confirmation of ownership, licensing terms, upload contents, or destination account. 6. The content becomes publicly available under MIT-0. This issue does not require a technical attacker; it arises from insufficient authorization controls around a consequential publication action. It can also be abused by a user who supplies third-party content that they are not authorized to license. ### Impact Assessment The workflow can publicly distribute the uploaded material under MIT-0 through the currently auth ...[truncated 394 chars]
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 (2)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The trigger phrases are broad and map to very common publishing requests, so the skill may auto-activate in situations where the user did not intend this specific high-privilege workflow. Because the skill performs local file operations and GUI automation, accidental invocation is more dangerous than for a read-only or low-privilege skill. The surrounding context increases risk because the triggers can launch actions that modify local files and interact with authenticated web sessions.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill goes beyond normal web publishing by deleting a local file and using GUI/clipboard automation to drive a system file picker. Those capabilities can affect local state outside the browser, and if the provided path is wrong, manipulated, or reused in a broader automation context, they could delete unintended files or interact with the wrong window/dialog. The context is a publishing workflow, which makes the behavior understandable, but it still expands the attack surface significantly compared with pure browser automation.

Content

No source excerpt is available for this finding.