Back to skill

Security audit

Paper2diagram

Security checks for vulnerabilities and agentic risk

Overview

This skill’s stated PDF-to-diagram purpose is coherent, but its actual executable workflow is not included and users are told to clone and run an unspecified external repository.

Install only if you can verify the external repository and its dependencies yourself. Use a trusted gateway, avoid confidential or unpublished PDFs unless your data-sharing obligations allow it, and keep API keys scoped and separate from installation steps where possible.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:57
Finding

Unverifiable External Repository and Dependency Installation

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 57–63
Vulnerability Type: Untrusted external source and insecure dependency installation
Risk Level: Medium

Vulnerable Code

bash
git clone <YOUR_REPO_URL> paper2diagram-agent
cd paper2diagram-agent
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env  # 按需填入 KEY

Technical Analysis

The supplied project contains only SKILL.md; it does not include the advertised Python workflow, requirements.txt, dependency lockfile, or integrity metadata. The installation instructions direct users to replace an unspecified repository placeholder and install dependencies defined by that external repository.

Consequently, the executable implementation and its dependency graph are outside the reviewed artifact and can change independently after this Skill is audited. In addition, Python package installation may execute package-controlled build hooks or installation logic. No dependency versions or cryptographic hashes are provided to ensure that users receive reviewed package contents.

This finding does not establish that a particular repository or package is malicious. The risk arises because the documentation establishes an unverified supply-chain path through which substituted, compromised, dependency-confused, or otherwise malicious components could be installed and executed.

Attack Path

  1. A user or agent follows the deployment instructions in SKILL.md.
  2. An attacker supplies a malicious value for YOUR_REPO_URL, compromises the selected repository, or introduces a malicious dependency through its requirements.txt.
  3. The user clones the repository without verifying its owner, commit, signature, or content.
  4. The user executes pip install -r requirements.txt.
  5. A malicious dependency executes code through package build or installation mechanisms, or the cloned workflow exe ...[truncated 1012 chars]
Remediation
View remediation

Remediation Suggestions

  1. Include the complete workflow implementation, dependency manifest, and relevant configuration in the auditable Skill package.
  2. Replace YOUR_REPO_URL with a specific official repository hosted under a verified project organization.
  3. Pin installation instructions to a reviewed immutable commit or signed release rather than a mutable branch.
  4. Provide a lockfile containing exact dependency versions and cryptographic hashes. Use hash-enforced installation, such as pip install --require-hashes.
  5. Review direct and transitive dependencies for dependency confusion, typosquatting, known vulnerabilities, and unexpected build scripts.
  6. Publish checksums or signatures for releases and document how users must verify them before installation.
  7. Run installation and document processing inside an isolated, least-privileged environment with restricted filesystem and network access.
  8. Do not expose API keys to package installation processes. Provide credentials only to the runtime process after dependencies and source contents have been verified.
  9. Add automated dependency scanning, provenance verification, and reproducible-build checks to the release process.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (3)

Credential Access

High
Category
Privilege Escalation
Confidence
60% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · SKILL.md (reported line 62)May include surrounding context.

python3 -m venv .venv source .venv/bin/activate pip install -r requirements.txt cp .env.example .env # 按需填入 KEY

text

2. 确认 `.env` 中的 KEY 与 gateway 地址正确可用。

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The markdown instructs the skill to generate summaries using a fixed Chinese-language '资深学术评审' style, and the example invocation is entirely in Chinese. This is a natural-language locale/style constraint presented as default behavior without offering the user a language choice or opt-in.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
93% confidence
Finding

The skill explicitly requires sending user-provided PDF content to externally configured LLM/image gateways, which is an external data transmission channel. Even if the gateway is user-controlled or trusted, PDFs may contain sensitive unpublished research, PHI, or proprietary data, so transmitting them off-host creates confidentiality and compliance risk.

Content

Scanner excerpt · SKILL.md (reported line 40)May include surrounding context.

md
- **需要网络访问**:连接到你自己配置的 LLM / 图像网关(例如 dongli gateway)
- **环境变量**(可以在 OpenClaw 的 `skills.entries.paper2diagram.env` 中配置,也可以在 shell 中设置):
  - `GEMINI_API_KEY`
  - `GEMINI_BASE_URL`(示例:`https://api.dongli.work/v1beta`)
  - `GEMINI_MODEL`(示例:`gemini-3-pro`)
  - `BANANA_PRO_API_KEY`
  - `BANANA_PRO_BASE_URL`(示例:`https://api.dongli.work`)