Back to skill

Security audit

浏览器代理助手

Security checks for vulnerabilities and agentic risk

Overview

This skill is a browser automation and scraping assistant, but it requests broad local command and file authority and documents proxy evasion, custom JavaScript, and detailed logging without enough limits or user controls.

Review before installing. Use this only in a sandboxed workspace with least-privilege credentials, avoid administrator execution, restrict target sites and writable paths, and do not use it on logged-in or sensitive pages unless logs and custom scripts are explicitly controlled and redacted.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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 (1)

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:13
Finding

Overprivileged Tool Access Violates Least-Privilege Boundaries

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:13-18; related unsafe privilege-elevation guidance at SKILL.md:364
Vulnerability Type: Excessive filesystem and command-execution privileges
Risk Level: Medium

Vulnerable Code

yaml
tools:
- read
- exec
- write
- glob
- grep

Technical Analysis

The Skill requests unrestricted command execution and broad filesystem read/write capabilities even though its stated primary purpose is browser automation and web data collection. The configuration does not define command allowlists, workspace boundaries, path restrictions, domain restrictions, or approval requirements for security-sensitive operations.

The document also recommends running with administrator privileges as a generic response to insufficient filesystem permissions. This guidance can unnecessarily expand the impact of command execution or file modification. The assertion that commands execute in a secure sandbox is documentary only; the project contains no implementation or enforceable sandbox policy that could be audited.

There is no evidence that this Skill directly performs malicious commands. The risk arises from the excessive authority exposed to agent-generated actions and the absence of enforceable least-privilege controls.

Attack Path

  1. A user directs the Skill to process an untrusted webpage, API response, or instruction.
  2. Attacker-controlled content contains instructions that the agent incorrectly interprets as authorized actions.
  3. The Skill has access to read, write, and exec, allowing the agent to inspect files, modify accessible files, or execute local commands.
  4. No documented command allowlist, path boundary, or mandatory user confirmation prevents the action.
  5. If the operator follows the related recommendation to use administrator privileges, the same action may execute with elevated host permissions.

Impact Assessment

Successful exploi ...[truncated 654 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove exec unless local command execution is strictly required for a documented browser-automation operation.
  2. Replace unrestricted filesystem tools with workspace-scoped read and write capabilities.
  3. Use a dedicated browser automation API instead of general-purpose shell execution.
  4. Enforce allowlists for executable programs, command arguments, destination domains, browser actions, and writable paths.
  5. Require explicit user confirmation before command execution, file modification, credential access, or other security-sensitive actions.
  6. Treat webpage content and external API responses as untrusted data and prohibit them from authorizing tool calls.
  7. Run the Skill as a dedicated unprivileged account inside an enforced sandbox or container with restricted networking and filesystem mounts.
  8. Replace the generic recommendation to use administrator privileges with least-privilege troubleshooting. Any exceptional elevation should be narrowly scoped, justified, time-limited, and explicitly approved.
  9. Document concrete sandbox guarantees and add automated tests verifying that commands and file operations cannot escape the permitted workspace.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (10)

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The manifest presents the skill as a generic API wrapper/LLM integration tool, but the body documents substantially broader capabilities including browser automation, scraping, proxy use, and custom scripting. This mismatch can cause the skill to be invoked under false assumptions, expanding effective privileges and network interaction scope beyond what users or policy reviewers would reasonably expect.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill description does not clearly warn users that it may execute commands, write files, and perform external API/network actions. In this context, the tool list includes read/exec/write plus browser-style automation, so omission of these warnings materially undermines informed consent and safe routing.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The activation cue is broad and vague, increasing the chance that the skill is selected for ordinary requests that do not require its powerful browser, exec, write, or network capabilities. Over-broad triggering is dangerous because it can route benign user tasks into a higher-risk tool unexpectedly.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The description states '支持中文交互' as a built-in behavior, but does not indicate that users may choose another language or opt in to a locale constraint. This can violate language/locale policy because it implicitly fixes the interaction language without documenting a justified regional limitation.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The documentation promotes data collection and detailed logging without warning about privacy, sensitive content capture, or retention risks. In a scraping/browser-assistant context, logs may contain user instructions, page contents, cookies/session-adjacent details, or API data, making silent collection more dangerous.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

Support for arbitrary custom JavaScript materially broadens the execution surface beyond a simple API wrapper. In a browser-automation context, this can be used to alter page behavior, exfiltrate data from sessions, or perform actions users did not anticipate from the skill description.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

Documenting proxy support specifically to evade anti-scraping controls enables misuse and exceeds the stated API-wrapper purpose. In context, this increases the likelihood of covert or policy-bypassing outbound activity against third-party sites, which is riskier than ordinary API access.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

Endorsing detailed operational logs without any minimization boundary creates a realistic risk of capturing sensitive user inputs and API interaction data. Because this skill handles browser automation, external requests, and possibly credentials or page content, unbounded logging can leak or retain sensitive material unnecessarily.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

Expanded guidance to store detailed logs and log files without content restrictions increases exposure of sensitive operational data at rest. In this skill, logs could accumulate instructions, scraped content, endpoint details, or error traces containing secrets, making later disclosure or misuse more likely.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

Troubleshooting guidance that suggests sending logs to console or files without safeguards encourages ad hoc exposure of sensitive data. Console output may be captured by surrounding systems, and file logs may persist beyond their need, especially in shared or multi-user environments.

Content

No source excerpt is available for this finding.