Back to skill

Security audit

转化路径优化虾

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed sales funnel and sales-script analysis helper that reads user-provided CRM files and produces local or user-requested reports, with only low-risk implementation cautions.

Before installing, use it only on CRM exports you are allowed to analyze, remove names, phone numbers, and other sensitive fields where possible, install dependencies in a virtual environment, and confirm before exporting reports or creating Feishu documents containing business data.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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

Note
Location
SKILL.md:71
Finding

Unpinned Third-Party Runtime Dependencies

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:71
Vulnerability Type: Unpinned third-party dependency installation
Risk Level: Low

Vulnerable Code

bash
pip install pandas openpyxl

Technical Analysis

The documented installation command retrieves the latest available versions of pandas, openpyxl, and their transitive dependencies without a lock file, exact version constraints, or package integrity hashes.

The package names are legitimate and there is no evidence that the project intentionally introduces malicious dependencies. Nevertheless, an installation performed at different times can resolve to different artifacts. A compromised upstream release, maliciously modified package-index response, or incompatible update could therefore introduce unreviewed code into the runtime environment.

Attack Path

  1. A user follows the dependency installation instructions in SKILL.md.
  2. pip resolves the packages and transitive dependencies available at installation time.
  3. A compromised or unexpectedly modified release is selected because no reviewed versions or hashes are enforced.
  4. Package installation or a subsequent import executes the affected dependency code under the user's account.

Impact Assessment

Exploitation would execute dependency code with the privileges of the user running pip or the analysis scripts. Depending on those privileges, the affected package could access CRM input data, files readable by the user, environment variables, and network resources.

The project itself does not demonstrate dependency confusion, typosquatting, use of an untrusted package source, or an existing compromised dependency. The finding concerns the absence of reproducible and integrity-verified dependency controls.

Remediation
View remediation

Remediation Suggestions

  1. Create a dependency lock file containing reviewed, exact versions of all direct and transitive dependencies.

  2. Generate and verify cryptographic hashes for every distribution.

  3. Install dependencies using hash enforcement, for example:

    bash
    python -m pip install --require-hashes -r requirements.txt
    
  4. Use an isolated virtual environment rather than installing packages globally or with administrative privileges.

  5. Configure an approved package index and disable unintended fallback indexes.

  6. Periodically scan and deliberately update the locked dependencies after compatibility and security review.

T09 · Insecure Skill Coding Practices

Note
Location
scripts/extract-scripts.py:112
Finding

Unsanitized CRM Content Written to Terminal Output

Content
View full analysis

Vulnerability Details

File Location: scripts/extract-scripts.py:112, scripts/extract-scripts.py:153, and scripts/extract-scripts.py:202
Vulnerability Type: Terminal control-sequence injection
Risk Level: Low

Vulnerable Code

The extraction command prints an untrusted script value directly:

python
print(f"   {script[:100]}{'...' if len(script) > 100 else ''}")

The ranking command uses the same unsafe output pattern:

python
print(f"   {script[:100]}{'...' if len(script) > 100 else ''}")

The recommendation command also emits the value without control-character sanitization:

python
print(f"   {script[:150]}{'...' if len(script) > 150 else ''}")

The script values originate from user-supplied CSV or Excel records and are not stripped of terminal control characters before being printed.

Technical Analysis

Limiting a string to 100 or 150 characters does not neutralize embedded ANSI escape sequences or other C0/C1 control characters. An attacker who can influence a CRM script field can insert sequences that change colors, move the cursor, clear or overwrite terminal content, create misleading hyperlinks, or manipulate terminal presentation.

The exact effect depends on the analyst's terminal emulator and its enabled features. No shell evaluates the value, so this is not command injection. The reviewed code also does not establish direct operating-system command execution through these print statements.

Attack Path

  1. An attacker or untrusted data source inserts terminal escape sequences into a sales-script field.
  2. The crafted record is exported to a CSV or Excel file.
  3. An analyst invokes the extract, rank, or recommend command on that file.
  4. The script reads the field and sends it directly to the analyst's terminal.
  5. The terminal interprets supported control sequences, allowing output to be hidden, altered, or visually spoofed.

Impac

...[truncated 483 chars]

Remediation
View remediation

Remediation Suggestions

  1. Sanitize every untrusted value before writing it to an interactive terminal.

  2. Remove or visibly escape C0 controls, DEL, C1 controls, and ANSI escape introducers.

  3. Preserve ordinary line breaks only when explicitly required and render other controls using escaped notation.

  4. Apply one centralized output-sanitization function consistently to all three commands. For example:

    python
    def sanitize_terminal(value: object) -> str:
        text = str(value)
        return "".join(
            ch if ch in "\t\n" or ord(ch) >= 0x20 and ord(ch) != 0x7F
            else f"\\x{ord(ch):02x}"
            for ch in text
        ).replace("\x1b", "\\x1b")
    
  5. Sanitize before truncation so a truncated value cannot retain an incomplete control sequence:

    python
    safe_script = sanitize_terminal(script)
    print(f"   {safe_script[:100]}{'...' if len(safe_script) > 100 else ''}")
    
  6. Consider offering JSON or another structured, noninteractive output mode for automated processing.

Vulnerability Patterns
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • 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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (7)

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

描述中有一部分与代码一致:代码确实能分析销售漏斗、计算各阶段转化率、找出低于行业基准的环节,并对比两个时间段表现,还能导出报告。但描述的核心范围明显更大,包含话术挖掘、客户画像驱动推荐、A/B 测试设计、销售培训手册生成等能力,而代码完全没有处理对话文本、客户画像、实验设计或内容生成功能。整体来看,代码只覆盖了“漏斗分析/时间段对比/报告导出”这一子集,因此声明对实际行为有明显夸大,属于描述与行为不匹配。

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
91% confidence
Finding

The skill advertises an output path that creates a Feishu document and the static analyzer detected file-write capability, but the manifest does not declare any tool scope or permissions. Undeclared write-capable behavior weakens reviewability and least-privilege controls, making it easier for a skill to create or modify artifacts without explicit operator awareness.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The natural-language description is written as a Chinese-only persona and workflow with no indication that users may choose another language or locale. Under the policy, forcing a specific language without opt-in is a natural-language policy violation unless the locale restriction is explicitly documented and justified.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

This file contains only Chinese-language templates and does not indicate that the skill is region-specific or that users can choose another language. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This code file contains natural-language usage instructions and output text exclusively in Chinese, which imposes a specific language/locale on all users without any opt-in or alternative. The policy explicitly flags language or locale constraints when the skill does not offer user choice or document a justified region-specific scope.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

The entire skill content is presented only in Chinese, starting with the title and continuing through all guidance, without any indication that users may choose another language. Under the stated policy, forcing a specific language without user opt-in can be a natural-language policy violation unless the locale restriction is clearly justified.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

This markdown file presents all instructions and framework details exclusively in Chinese, and there is no indication that the skill is region-specific or that users can opt into another language. Under the natural-language policy rules, forcing a specific language without user choice can be a policy violation.

Content

No source excerpt is available for this finding.