Back to skill

Security audit

YouTube 公开视频元数据中文化

Security checks for vulnerabilities and agentic risk

Overview

The translation workflow is limited, but conflicting instructions leave unclear whether you control which browser profile the agent uses.

Require one explicitly selected browserId throughout the workflow, with separate authorization for creating environments or installing software. Prefer a dedicated profile and local draft output. Review and pin installation versions where possible. The Review verdict concerns unclear profile-selection authority; the artifacts do not establish malicious intent.

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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (7)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The supplied command performs skill installation, not video-page reading or metadata localization. Installation is a materially different operation with local environment effects, rather than an implementation detail of the stated browser-based review workflow. The chunk does not establish what the installed skill subsequently does, so no additional downstream behavior is inferred.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
93% confidence
Finding

The installation command invokes an unpinned npm CLI through npx and installs skill content without an immutable revision. A compromised or malicious future release could execute code with the installing user's permissions or introduce adversarial skill instructions; the read-only browser workflow does not protect the installation stage. The scanner's MCP-server classification is inaccurate, but the underlying supply-chain exposure is real.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · README.en.md (reported line 40)May include surrounding context.

md
Bridge and browser extension on Apple Silicon Macs and Windows, so no separate Bridge installation is needed there. Ordinary browsers and other systems should follow the current Easy WebBridge setup described in the official guide.

For a browser task the user has authorized through this Skill, check and reuse a suitable existing environment first. If a new one is genuinely needed, create one unique environment for this task with safe defaults, read back its actual environment ID, and start it. Do not ask the user to reconfirm or supply an environment name or default parameters. After confirming the correct account and `browserId`, continue the user's current task instead of stopping at setup instructions; do not ask for approval at each necessary step. Pause only for CAPTCHA/account security checks, real ambiguity about the target account or authorization, or missing local-control capability. At completion, close only the task tab group created for this run and report results and limi

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
93% confidence
Finding

Although this command invokes a skill installer rather than an MCP server, npx skills add ... can download and execute an unpinned version of the skills npm package. A compromised future release could execute attacker-controlled code with the invoking user's permissions. Its placement in installation documentation makes execution plausible, but the supplied content provides no evidence of an actual compromise or malicious intent.

Content

No source excerpt is available for this finding.

The workflow permits filling draft fields despite the manifest's prohibition on modifying backend content.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

The manifest at L03 limits the skill to reading public metadata and producing review drafts, explicitly stating that backend content is not modified; L20 reinforces this restriction. L64 nevertheless authorizes filling draft fields when requested, introducing an editing operation outside that declared scope even though final save or submission is prohibited.

Content

No source excerpt is available for this finding.

The continuation instructions allow autonomous browser-environment selection and creation instead of requiring the user-designated environment.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The continuation section permits autonomous environment selection or creation despite earlier requirements for an explicitly user-designated browserId. This conflicting authorization rule could cause the agent to inspect or operate in an unintended authenticated browser profile; the later account-check requirement reduces risk but does not clearly restore user control over environment selection.

Content

No source excerpt is available for this finding.

Scope Creep

Low
Category
Excessive Agency
Confidence
70% confidence
Finding

Skill's behavior or capabilities extend beyond its stated purpose. Scope creep allows an agent to perform actions unrelated to its documented functionality, increasing the attack surface.

Content

Scanner excerpt · LICENSE (reported line 7)May include surrounding context.

text
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.