The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script (scripts/feedback-server.py) using uv run. This script operates a one-shot HTTP server on 127.0.0.1 to receive user feedback from the browser interface and save it to a file for the agent to read.
[EXTERNAL_DOWNLOADS]: The assets/review.html file references external stylesheets and fonts from Google's official domains (fonts.googleapis.com and fonts.gstatic.com). These are well-known technology services used for UI rendering and do not contribute to verdict escalation.
[PROMPT_INJECTION]: The skill renders scraped HTML content within an iframe using sandbox="allow-same-origin". Because the review page and the scraped pages are opened as sibling file:// URIs, scripts within the scraped content could potentially access the parent page's JavaScript state, manipulate the review results, or automate the feedback submission. This represents an indirect prompt injection surface.
Ingestion points: Untrusted scraped HTML files are loaded into an iframe; user feedback is captured in feedback.txt via a local server and then read by the agent.
Boundary markers: No explicit boundary markers or instructions are used to separate user-provided feedback from potential script-injected content in the resulting feedback file.
Capability inventory: The agent has access to Bash, Read, and Write tools across all scripts to process the data.
Sanitization: While the JavaScript interface uses escapeHtml for rendering structured data values, the iframe renders raw HTML content without strict script isolation, allowing for potential interaction with the feedback mechanism.

scrape-review-schema