The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill performs network operations to scrape ranking and book data from several external domains including qidian.com, fanqienovel.com, jjwxc.net, qimao.com, and ciweimao.com through a variety of scraper scripts (e.g., scripts/qidian-rank-scraper.js).
[COMMAND_EXECUTION]: The utility script scripts/cdp-utils.js invokes shell commands via execSync to communicate with the agent-browser tool. This is the intended functional path for performing browser-based scraping within the platform.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data (novel titles, descriptions, and tags) from public websites and incorporates this content into the agent's context to generate market reports and decision documents. \n
Ingestion points: Untrusted novel metadata and descriptions are scraped in scripts such as scripts/qidian-rank-scraper.js (via descText) and scripts/fanqie-rank-scraper.js (via info.desc).\n
Boundary markers: There are no explicit delimiters or instructions to the agent to disregard potential commands found within the scraped content.\n
Capability inventory: The skill has the capability to write local files (fs.writeFileSync) and execute browser-based commands.\n
Sanitization: Basic string normalization and truncation logic exists (e.g., cleanDesc in fanqie-rank-scraper.js), but no semantic filtering is applied to detect or strip instruction-like text.

story-long-scan