paper-2-web
Fail
Audited by Gen Agent Trust Hub on Jul 13, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTIONNO_CODE
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill instructs the agent to clone an external, untrusted repository from an unverified source (https://github.com/YuhangChen1/Paper2All.git) and execute Python scripts such as 'pipeline_all.py' and 'pipeline_light.py'. This architecture allows for the execution of arbitrary remote code on the user's environment.
- [EXTERNAL_DOWNLOADS]: Installation involves downloading source code and installing unverified dependencies via 'pip install -r requirements.txt' from an untrusted third-party repository. This practice lacks integrity verification and introduces significant supply chain risks.
- [NO_CODE]: The skill does not contain any executable scripts or local code logic. It functions solely as a trigger to download and execute external code, which is an insecure design pattern.
- [COMMAND_EXECUTION]: The skill makes extensive use of the 'Bash' tool to perform high-risk operations, including repository cloning, environment management with 'conda', and running complex system-level transformation pipelines.
- [PROMPT_INJECTION]: The skill is highly susceptible to indirect prompt injection due to its processing of untrusted academic papers (LaTeX and PDF). It lacks the necessary security controls for this attack surface:
- Ingestion points: The agent reads raw LaTeX source and PDF files provided by the user.
- Boundary markers: Absent; there are no delimiters or instructions to ignore embedded commands in the source documents.
- Capability inventory: The skill has 'Bash' access, file write permissions, and potential network access via the Google Search API.
- Sanitization: No sanitization, validation, or filtering of the extracted content is mentioned before it is processed by the LLM or executed as commands.
- [PROMPT_INJECTION]: Deceptive metadata is present in the skill; the 'skill-author' field in SKILL.md is listed as 'K-Dense Inc.', which does not match the provided developer context of 'DenDen047'. This discrepancy can mislead users about the source and trustworthiness of the skill.
Recommendations
- AI detected serious security threats
Audit Metadata