The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill performs runtime installation of the cloudscraper Python package if it is not detected on the system using pip install --user in SKILL.md.
[COMMAND_EXECUTION]: The skill modifies system and application preferences, specifically enabling the AllowJavaScriptFromAppleEvents and IncludeDevelopMenu settings in Safari to facilitate browser automation in references/oceanofpdf-download.md.
[COMMAND_EXECUTION]: The skill utilizes osascript to execute dynamically generated JavaScript code within the user's Safari browser to handle form submissions on external websites in references/oceanofpdf-js-via-tempfile.md.
[COMMAND_EXECUTION]: The skill executes various system discovery commands such as networksetup and scutil to extract proxy configurations from the host machine in SKILL.md.
[DATA_EXFILTRATION]: The skill programmatically identifies local proxy servers (e.g., Clash Verge) and exports their address/port to environment variables for use in all subsequent network operations in SKILL.md.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests and processes untrusted metadata (titles, authors, and descriptions) from numerous external sources such as Libgen, Douban, and Open Library to influence agent decision-making.
Ingestion points: scripts/ebook_fetch.py (Open Library, Google Books), scripts/libgen_fetch.py (Libgen), and scripts/resolve_metadata.py (Douban).
Boundary markers: None identified in the instructional prompts.
Capability inventory: File system writes (downloads), network operations, and shell command execution via the terminal tool.
Sanitization: The clean_remote function in scripts/ebook_fetch.py restricts string length and removes certain control characters, providing a basic layer of protection.

ebook-download