newsletter-designer
Pass
Audited by Gen Agent Trust Hub on Mar 23, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local maintenance and generation scripts using shell commands.
- Executes
python3 scripts/flux_image.pyto generate hero images. - Executes
node scripts/export_pdf.jsto render the final PDF. - Invokes
npm installwithin the scripts directory to manage dependencies. - [EXTERNAL_DOWNLOADS]: The skill fetches external resources during its execution flow.
- The
flux_image.pyscript downloads generated image files from remote URLs provided by image generation providers. - Downloads the
puppeteerpackage and its required browser binaries via NPM. - [DATA_EXFILTRATION]: The skill transmits data to external service providers to fulfill its primary purpose.
- Transmits user-provided prompts and API keys (
BFL_API_KEY,OPENROUTER_API_KEY) to Black Forest Labs (api.bfl.ml) and OpenRouter (openrouter.ai) endpoints. - Extracts content from user-provided URLs during the brand extraction phase.
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8).
- Ingestion points: Untrusted data enters the agent context via the 'Client website URL' (Phase 1) and raw 'Newsletter content' (Phase 0) provided by users.
- Boundary markers: No explicit delimiters or instructions are used to isolate the scraped website content or user text from the agent's core instructions.
- Capability inventory: The skill possesses the ability to execute shell commands (
python3,node), perform network requests (urllib), and write files to the local disk (./output/). - Sanitization: There is no evidence of sanitization or filtering of the external content before it is interpolated into the HTML template or processed by the AI model. An attacker-controlled website could include instructions that influence the agent's behavior during the design process.
- [COMMAND_EXECUTION]: The PDF generation script (
export_pdf.js) launches a headless browser with the--no-sandboxand--disable-setuid-sandboxflags. This reduces the security isolation of the browser process while it renders HTML that may contain untrusted content derived from scraped websites.
Audit Metadata