Back to skill

Security audit

多视角投研分析

Security checks for vulnerabilities and agentic risk

Overview

The skill is an A-share research helper, but it overstates implemented security controls and gives broad command, dependency, API-key, and administrator guidance that users should review carefully.

Install only if you are comfortable reviewing each command before it runs, configuring third-party finance API keys yourself, and treating the listed security protections as claims rather than verified controls. Do not run the dependency installation or troubleshooting steps with administrator privileges unless you have independently narrowed the exact command and file permissions needed.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (2)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:275
Finding

Unpinned Third-Party Dependency Installation

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, line 275
Vulnerability Type: Unpinned and unverifiable third-party dependencies
Risk Level: Medium

Vulnerable Code Snippet:

markdown
- Dependency installation command: `pip install numpy pandas scikit-learn tensorflow`

Technical Analysis

The installation command does not constrain package versions, verify package hashes, use a lockfile, or specify an approved package repository. Consequently, the packages and their transitive dependencies can change between installations without any modification to the reviewed Skill.

Python package installation can execute package build and installation logic. If a package release, transitive dependency, or configured package index is compromised, following this instruction may execute attacker-controlled code with the privileges of the user running pip.

The named packages are legitimate and there is no evidence that they are currently malicious. The vulnerability arises from the unsafe, mutable dependency-resolution process rather than from a confirmed malicious package.

Attack Path

  1. A user or Agent follows the dependency installation instruction.
  2. pip queries the environment's configured package indexes and resolves the latest compatible versions.
  3. A compromised package release, transitive dependency, or malicious package index supplies attacker-controlled package content.
  4. Package build or installation hooks execute attacker-controlled code.
  5. The code gains the permissions of the account running pip; if installation is combined with the document's administrator-execution advice, the impact may extend to elevated system privileges.

Impact Assessment

Successful exploitation could permit arbitrary code execution under the installing user's account. This may expose files, environment variables, configured API keys, financial datasets, and other resources available to that account. ...[truncated 238 chars]

Remediation
View remediation

Remediation Suggestions

  1. Replace the inline installation command with a reviewed dependency manifest containing exact versions.
  2. Generate and enforce cryptographic hashes, for example by using a hash-locked requirements file and pip install --require-hashes.
  3. Pin all transitive dependencies through a reproducible lockfile.
  4. Explicitly require installation from an approved package repository and reject untrusted extra indexes.
  5. Require installation in a dedicated, unprivileged virtual environment or container.
  6. Add automated dependency vulnerability and provenance scanning.
  7. Document a controlled process for reviewing and updating pinned versions.

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:325
Finding

Generic Administrator-Privilege Escalation Guidance

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 325 and 347
Vulnerability Type: Unnecessary privilege escalation and violation of least privilege
Risk Level: Medium

Vulnerable Code Snippet:

The following table row appears twice in the source, at lines 325 and 347. This is an English translation of the source text:

markdown
| Insufficient permissions | The current user lacks read/write permissions | Check file permissions and run as administrator |

Technical Analysis

The troubleshooting guidance recommends running an operation as administrator whenever a generic file-permission error occurs. It does not identify the specific file or directory requiring access, explain the minimum permissions needed, distinguish read access from write access, or restrict which command may be elevated.

This breaks the principle of least privilege. Permission errors commonly indicate incorrect paths, unsafe output locations, ownership problems, or an improperly configured environment; they should not automatically result in administrator-level execution.

The risk is amplified because the same document also instructs users to install mutable third-party dependencies and discusses unspecified command execution. If either operation is rerun with administrator privileges, any malicious dependency or unsafe command would inherit elevated access.

Attack Path

  1. A workflow encounters a file read/write permission error.
  2. The user follows the Skill's troubleshooting advice and reruns the relevant operation as administrator.
  3. The operation executes an untrusted dependency, unsafe command, or compromised process with elevated privileges.
  4. That process modifies or reads system resources that were inaccessible to the original unprivileged account.
  5. Depending on the operating system and elevated token provided, the process may compromise system-wide data or configuration.

Impact Assessment

Expl ...[truncated 533 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove the generic recommendation to run operations as administrator.
  2. Identify the exact file or directory involved and determine whether read, write, or execute access is genuinely required.
  3. Recommend narrowly scoped ownership or permission changes only for application-specific paths.
  4. Require execution through a dedicated unprivileged account, virtual environment, or sandbox.
  5. Prohibit elevated dependency installation and model execution.
  6. If elevation is unavoidable, document the exact reviewed command, required resource, expected changes, and rollback procedure.
  7. Apply the same correction to both duplicated troubleshooting entries at lines 325 and 347.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • 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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (6)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
81% confidence
Finding

The visible name, summary, and descriptive instructions are entirely Chinese and scoped to A股 users, but the file does not state that this locale restriction is optional or provide a language choice. Under the policy, a language or locale constraint should be opt-in or clearly justified as region-specific.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The document claims that risky code and external dependencies were removed and that security/stability were enhanced, yet the skill still declares the exec tool. This creates a misleading trust signal: operators may assume the skill cannot execute commands when it still has command-execution capability, increasing the chance of unsafe deployment or over-trusting its behavior.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The activation text is broad enough to match generic analysis-related requests, which can cause the skill to trigger outside a narrowly intended finance context. Because the skill also declares exec and discusses external APIs, over-broad invocation increases the chance of unnecessary tool exposure, unexpected data access, or user confusion about what actions may be taken.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The Markdown file asserts that API key management, encryption, logging protections, input validation, access control, and vulnerability scanning are implemented, but the file contains no code or enforceable mechanism providing those controls. These unsupported security claims can mislead users into storing secrets or using the skill in sensitive workflows under false assumptions about protection.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill references external market/news data and later requires API keys, but does not clearly warn users up front that third-party data access and secret configuration are required. This can lead users to provide credentials or permit outbound data usage without informed consent, which is especially sensitive in finance workflows.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

The later sections instruct users to train models, evaluate them, and deploy them with example Python code, creating the impression that this skill implements ML training functionality. Yet L357-L359 state the skill is 'MD(纯Markdown指令,通过自然语言驱动Agent完成操作)', so the document's implementation-oriented examples contradict the actual nature of the artifact.

Content

No source excerpt is available for this finding.