web-access
Warn
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the user to visit 'https://treg.to/llms.txt' to install the 'treg' CLI tool if it is not already configured. This promotes the installation of third-party software from an external, non-whitelisted domain.
- [COMMAND_EXECUTION]: The skill extensively utilizes the 'treg' command-line interface to perform web searches, page extractions, and site mapping (e.g.,
treg call treg.web.search). - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from untrusted external sources, including arbitrary URLs and social media platforms.
- Ingestion points: Web page content fetched via
treg.web.extract, search result snippets, and social media data (posts, comments, and reviews from X, LinkedIn, Reddit, TikTok, etc.). - Boundary markers: There are no explicit instructions or delimiters defined to isolate the fetched external content from the agent's core instructions.
- Capability inventory: The agent can execute CLI commands, map entire websites, and perform structured data extractions based on the external content.
- Sanitization: The instructions do not define any sanitization, filtering, or validation steps for the content retrieved from external vendors before it is processed.
Audit Metadata