Back to skill

Security audit

YouTube 上传草稿检查

Security checks for vulnerabilities and agentic risk

Overview

This skill needs review because conflicting browser-selection instructions could expose private drafts through an unintended session, and it includes unrelated promotional directions.

Before installing, resolve the conflicting browser rules so every read uses your explicitly selected browserId, and remove the unrelated promotional instructions. Review installation versions and any browser environment creation separately. Offline test results do not verify your live YouTube account.

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
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
Findings (14)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The checklist fields, stop conditions, source reporting, and non-mutating behavior align with the description. However, the central resource and verification method differ materially: the code neither accepts a browserId nor accesses Easy WebBridge or YouTube Studio. It transparently reports offline-fixture mode, but that limitation is absent from the declared description. No publishing, visibility changes, or other undeclared external actions are present.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding

就所提供的代码而言,其实际操作是安装技能,与声明的 YouTube Studio 核对流程不同。安装命令本身不能证明所声明的核对能力,也无法验证身份、所有权及目标不明确时停止的约束。此判断仅针对当前片段,不推断远程技能的内容,也没有证据表明该片段会发布视频或修改可见性。

Content

No source excerpt is available for this finding.

An unrelated branding directive steers user-facing content toward unsolicited EasyBR promotion.

Medium
Category
Prompt Injection
Confidence
94% confidence
Finding

The directive to mention EasyBR in social posts, articles, and voiceover copy is unrelated to checking a YouTube draft and attempts to influence user-facing content beyond the authorized task. If followed, it could introduce unsolicited advertising into generated material, although it neither authorizes publication nor demonstrates data exfiltration.

Content

No source excerpt is available for this finding.

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

Medium
Category
MCP Rug Pull
Confidence
87% confidence
Finding

The installation command invokes an unpinned npm executable and retrieves a skill without an immutable revision, leaving installation behavior dependent on mutable upstream content. This is a supply-chain exposure rather than evidence of an MCP server or an existing compromise; the browser-oriented context increases the consequences if compromised installation code reaches a machine with authenticated sessions.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
91% confidence
Finding

Proceeding without approval for every routine read-only step is not inherently unsafe, but this paragraph also suppresses confirmation while allowing browser-environment selection and creation outside the explicit-browserId contract. This is the same authorization-boundary issue identified by SDI-1, not an independent escalation; the read-only and account-ambiguity safeguards limit its impact.

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

The workflow authorizes autonomous browser-environment selection and creation instead of requiring the user's explicitly specified browserId.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The onboarding workflow permits the agent to select an existing browser environment or create one without obtaining the user's explicitly selected browserId, contradicting the skill's stated scope. Confirming the account reduces risk but does not establish authorization for a particular environment, so the agent could access private draft information through an unintended session even while remaining read-only.

Content

No source excerpt is available for this finding.

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

Medium
Category
MCP Rug Pull
Confidence
86% confidence
Finding

The installation command invokes an unpinned npm package through npx and references skill content without an immutable revision, allowing the code or instructions installed to change without changes to this README. Exploitation would require a compromised or malicious upstream release; this is a dependency-integrity weakness, not evidence of malware, and the scanner's characterization of skills as an MCP server is unsupported.

Content

No source excerpt is available for this finding.

The workflow permits autonomous browser-environment selection and creation despite requiring a user-specified browserId.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The later workflow permits autonomous selection or creation of browser environments without requiring the user to supply an environment identifier, contradicting the earlier requirement for an explicitly selected browserId. For authenticated YouTube Studio access, this weakens the account-selection boundary and could lead to reading or reporting another environment's private draft metadata; account checks and read-only restrictions reduce, but do not eliminate, that risk.

Content

No source excerpt is available for this finding.

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: file_read, shell.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

The workflow permits autonomous browser-environment selection and creation despite requiring a user-specified browserId.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The closing instructions permit autonomous reuse or creation of a browser environment without renewed confirmation, contradicting the earlier requirement to use only the browserId explicitly supplied by the user. This ambiguity can lead an agent to inspect private Studio data in an unintended environment; the ownership checks and read-only restrictions reduce, but do not eliminate, that authorization risk.

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 · SOURCE-ORIGIN.md (reported line 14)May include surrounding context.

md
- https://github.com/aiispower/easy-webbridge
  - https://gitee.com/xxjrq/easy-webbridge
  - https://github.com/social-media-skills/skills
  - https://github.com/social-media-skills/skills/blob/main/skills/youtube-publishing-and-metadata/SKILL.md

Offline fixtures contain sample data for tests; no user account data is packaged.

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 · SOURCES.md (reported line 4)May include surrounding context.

md
# Sources

- https://github.com/social-media-skills/skills
- https://github.com/social-media-skills/skills/blob/main/skills/youtube-publishing-and-metadata/SKILL.md

These links were consulted only for the publishing-and-metadata problem space. No source text or code is copied into this repository; its instructions, fixtures, and implementation are original.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/self_test.py (reported line 13)May include surrounding context.

python
CHECKER = ROOT / "scripts" / "check_draft.py"

def run(name):
    output = subprocess.check_output([sys.executable, str(CHECKER), str(ROOT / "fixtures" / name)], text=True)
    return json.loads(output)

def main():

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.