The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads an executable CLI binary from the vendor's official GitHub organization to provide hotel search and booking functionality.
Evidence: scripts/install.py fetches assets from github.com/RollingGo-AI/oauth-hotel-cli-overseas and fallback URLs pointing to the same repository.
[COMMAND_EXECUTION]: The skill executes shell commands to perform installation and interact with the hotel booking CLI.
Evidence: scripts/install.py executes npm install for the Node-based version of the tool. scripts/rgg.js and scripts/rgg.py act as wrappers that execute the rgg binary using spawn and subprocess.run respectively.
[PRIVILEGE_ESCALATION]: The installation script modifies file permissions to enable execution of the downloaded CLI tool on Unix-based systems.
Evidence: scripts/install.py calls os.chmod(dest_path, 0o755) to set executable bits for the downloaded binary.
[INDIRECT_PROMPT_INJECTION]: The skill ingests hotel information from an external CLI tool, which represents a potential surface for indirect prompt injection via untrusted data.
Ingestion points: Hotel names, descriptions, and tag information retrieved via the search-hotels and hotel-detail subcommands of the rgg CLI.
Boundary markers: The skill instructions provide structured Markdown templates in SKILL.md to display hotel data to the user, though explicit "ignore embedded instructions" delimiters are not present.
Capability inventory: The skill possesses the capability to execute binaries (rgg) and perform network operations via those binaries, as managed by the proxy wrappers in scripts/rgg.js and scripts/rgg.py.
Sanitization: The instructions include strict output specifications that forbid the display of technical details (IDs, raw JSON, CLI command logs) to the user, reducing the risk of data contamination affecting the agent's logic.

rollinggo-hotel-booking