Back to skill

Security audit

LLM数据自动化

Security checks for vulnerabilities and agentic risk

Overview

This skill is mostly a coherent data-automation helper, but its setup guidance includes unsafe remote script execution and under-disclosed handling of potentially sensitive data.

Review this skill carefully before installing. Avoid copying the Ollama curl-to-sh command; use a verified installation method instead. Treat generated scripts as code that can read and overwrite files, run them on copies or in a sandbox first, and avoid sending confidential datasets or prompts to the recommended cloud API unless its privacy and retention terms are acceptable.

Vulnerability Patterns
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

T03 · Remote Payload Retrieval and Execution

Error
Location
SKILL.md:74
Finding
Unverified Remote Installer Is Piped Directly into a Shell## Vulnerability Details **File Location**: `SKILL.md`, line 74 **Vulnerability Type**: Remote payload retrieval and execution **Risk Level**: High **Complete Code Snippet**: ```bash curl -fsSL https://ollama.com/install.sh | sh ``` ### Technical Analysis The installation instructions download a mutable remote script and execute it immediately through `sh`. The command provides no version pinning, cryptographic checksum verification, publisher-signature validation, or opportunity to inspect the downloaded content before execution. As a result, the effective payload is not contained in the audited Skill and can change after the package has been reviewed. Trust is delegated entirely to the remote domain, its hosting infrastructure, DNS resolution, certificate trust chain, and upstream release process. Ollama is presented as an optional local LLM provider, so direct remote-script execution is not necessary for the Skill's declared data-processing functionality. A verified package-manager installation or a separately downloaded, version-pinned, authenticated artifact would provide the optional dependency without exposing an unrestricted remote code-execution channel. ### Attack Path 1. A user follows the installation instructions in `SKILL.md`. 2. `curl` retrieves the current response from `https://ollama.com/install.sh`. 3. The response is passed directly to `sh` without local inspection or integrity verification. 4. If the remote origin, distribution infrastructure, DNS/TLS trust path, or installer content is compromised, attacker-controlled shell commands execute locally. 5. Those commands operate with the privileges of the user running the command and can invoke privilege-elevation mechanisms if the user authorizes them. ### Impact Assessment A malicious or compromised installer can execute arbitrary commands with the invoking user's privileges. This may permit access to or modification of user-readable files, theft ...[truncated 613 chars]
Remediation
## Remediation Suggestions 1. Remove the `curl | sh` installation command from the Skill documentation. 2. Prefer an official, authenticated operating-system package manager and pin the expected package version where supported. 3. If a standalone installer is required: - Download it to a local file without executing it. - Pin an immutable release URL and explicit version. - Obtain the expected SHA-256 digest through a separately authenticated channel. - Verify the digest and the publisher's cryptographic signature. - Inspect the script before execution. - Run it as an unprivileged user unless a documented installation step strictly requires elevation. 4. Document the exact filesystem and system changes performed by the installer. 5. Treat Ollama as an optional external prerequisite rather than automatically installing it as part of the Skill workflow. 6. For stronger supply-chain controls, distribute a reproducible, signed package and verify its provenance or attestation before installation.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (6)

External Script Fetching

High
Category
Supply Chain
Content
```bash
  # 安装Ollama
  curl -fsSL https://ollama.com/install.sh | sh
  
  # 下载模型
  ollama pull mistral
Confidence
98% confidence
Finding
The skill instructs users to pipe a remotely fetched script directly into the shell using 'curl ... | sh', which executes unreviewed code from the network with the user's privileges. If the source is compromised, intercepted, or changed unexpectedly, this can lead to arbitrary code execution and full system compromise.

Chaining Abuse

High
Category
Tool Misuse
Content
```bash
  # 安装Ollama
  curl -fsSL https://ollama.com/install.sh | sh
  
  # 下载模型
  ollama pull mistral
Confidence
97% confidence
Finding
The shell pipe into 'sh' creates a high-risk command chain that immediately executes external content without an inspection step, amplifying supply-chain and command-execution risk. In a skill aimed at non-expert users, this is more dangerous because readers may copy-paste the command verbatim without understanding the trust implications.

Missing User Warnings

Medium
Confidence
89% confidence
Finding
The documentation encourages users to generate executable Python and 'directly run' it without warning that the produced code can read local files, write outputs, overwrite existing data, or perform unintended file operations. Because this skill is specifically about automating data processing, users may run generated scripts against valuable local datasets and unintentionally damage, leak, or corrupt them.

Missing User Warnings

Medium
Confidence
91% confidence
Finding
The skill recommends use of a third-party cloud API but does not warn that user prompts, file-derived content, or potentially sensitive business data may be transmitted off-device to an external service. In a data-automation context, users are especially likely to process confidential spreadsheets, PDFs, or operational data, so the missing disclosure materially increases privacy and compliance risk.

Vague Triggers

Medium
Confidence
90% confidence
Finding
The instruction activates on broad user intents like 'convert or extract data' without narrowing allowed file types, trusted sources, or operation boundaries. In a skill that generates Python/Pandas workflows and document extraction pipelines, this can cause the agent to over-trigger on ordinary requests and perform higher-risk data handling actions than the user explicitly scoped, increasing the chance of unsafe file access, unintended processing of sensitive data, or execution-adjacent behavior through downstream code-generation patterns.

Natural-Language Policy Violations

Low
Confidence
94% confidence
Finding
文档写明“直接用中文描述你的数据处理需求”,属于自然语言层面的语言限定,但未说明这是可选项,也未给出其他语言支持或用户选择。这构成潜在的语言/locale 策略问题。

Static analysis

No suspicious patterns detected.