The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection by ingesting search results from an external API and displaying them to the agent.\n
Ingestion points: Search results containing titles, author names, and content snippets are fetched from the Toutiao API in scripts/search.py.\n
Boundary markers: Untrusted content is not separated from agent instructions by specific delimiters or boundary warnings.\n
Capability inventory: The skill has the ability to write local files (CSV/HTML), perform network requests, execute shell commands (to open a browser), and run a persistent local HTTP server.\n
Sanitization: While the HTML template (assets/report_template.html) employs basic HTML escaping, there is no validation or filtering of the content before it is processed by the agent.\n- [EXTERNAL_DOWNLOADS]: The skill requires the requests library and uses the alipay-bot CLI tool for payment processing, which is installed via NPM from the @alipay/agent-payment package.\n- [COMMAND_EXECUTION]: The scripts/search.py script uses subprocess.run to execute the system open command, which launches the generated report in the default web browser.\n- [DYNAMIC_EXECUTION]: The skill generates interactive HTML reports at runtime and starts a local HTTP server (HTTPServer) on 127.0.0.1. This server proxies API requests to redfox.hk to handle in-page search functionality while bypassing cross-origin restrictions.\n- [DATA_EXPOSURE]: The skill reads the REDFOX_API_KEY from ~/.qoder/apis/redfox.json or environment variables for authentication. This follows the vendor's standard practice for credential management.

toutiao-search