x-articles
Pass
Audited by Gen Agent Trust Hub on Mar 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The script
scripts/publish-article.shinvokes theagent-browserCLI tool to perform automated actions within a web browser, including navigation and UI interaction. - [COMMAND_EXECUTION]: The script
scripts/publish-article.shuses thepbcopycommand to move data from the article file to the system clipboard. - [EXTERNAL_DOWNLOADS]: The skill requires the
agent-browserpackage, which must be installed from a standard public registry (NPM). - [PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection (Category 8) due to the way it handles external content.
- Ingestion points: Untrusted article content is read from local files by
scripts/format-for-x.shandscripts/publish-article.sh. - Boundary markers: The processing flow lacks delimiters or specific instructions to ignore embedded agent commands within the text.
- Capability inventory: The skill uses
agent-browserto control a logged-in X (Twitter) account and can execute JavaScript via theevaluatemethod to interact with the page DOM. - Sanitization: There is no sanitization or verification logic to ensure the article content does not contain instructions designed to manipulate the agent's browser automation behavior.
Audit Metadata