Back to skill

Security audit

B站简介资料链接索引

Security checks for vulnerabilities and agentic risk

Overview

This skill provides a bounded workflow for indexing public video-description links, with ordinary installation cautions and an offline parser limitation.

Review the external installer and browser dependencies before installation. Keep operations bound to your chosen browserId and explicitly name any authorized draft destination. Treat offline parser results as unverified links, not proof that the author visibly supplied them.

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)

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

Medium
Category
MCP Rug Pull
Confidence
87% confidence
Finding

The installation command uses an unpinned skills npm package and does not specify an immutable revision of the skill repository. A compromised or malicious future release could execute installer code with the invoking user's privileges or introduce adversarial skill instructions; however, the static finding incorrectly describes this as an MCP server reference, and there is no evidence of an actual compromise.

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 51)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
91% confidence
Finding

The installation command invokes an unpinned npm CLI through npx skills, which can download and execute a changing package version with the invoking user's permissions. A compromised release could execute malicious code during installation; the browser workflow's read-only restrictions would not constrain that installer. Although the finding incorrectly describes this as an MCP server reference, the underlying dependency-pinning concern is valid, with no evidence of malicious intent in the supplied file.

Content

No source excerpt is available for this finding.

The workflow permits writing into a browser draft interface despite the manifest's read-only scope.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The manifest at L03 limits the skill to reading visible descriptions, checking public links, and delivering an index draft. L44 additionally authorizes filling a draft interface, which modifies browser application state rather than merely returning the index. Explicit user authorization and the prohibition on final submission reduce the risk but do not make this capability part of the declared scope.

Content

No source excerpt is available for this finding.

The continuation workflow allows autonomous environment selection and creation instead of requiring the user-specified browserId.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

The continuation section permits autonomous reuse or creation of a browser environment despite earlier requirements for a user-specified browserId and explicit selection when multiple environments exist. This ambiguity can cause an agent to treat general task authorization as permission to select a different authenticated environment, creating cross-account access or unintended local state changes; the account-verification requirement and read-only default reduce, but do not eliminate, that risk.

Content

No source excerpt is available for this finding.

L11–L16 index every anchor in the supplied HTML as visible source material instead of limiting extraction to the video's visible description.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

The parser extracts every matching anchor from the supplied HTML without verifying that it belongs to the video's description or is actually visible, allowing hidden, unrelated, or attacker-inserted links to be attributed to the specified video. This violates the skill's core provenance constraint and could contaminate the delivered source index with misleading or malicious destinations, although the script itself neither visits links nor executes their content.

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.