The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the Python subprocess module to orchestrate its internal components and manage the runtime environment. Scripts such as scripts/core/scrape_all_sources.py and scripts/management/refresh_index.py run other skill-resident scripts as subprocesses. Additionally, scripts/setup/setup_dependencies.py executes system package managers (winget, choco, apt, pacman) and pip to ensure required and optional libraries are installed.\n- [EXTERNAL_DOWNLOADS]: The skill fetches documentation data from geminicli.com using the llms.txt format. It also performs downloads from official package registries like PyPI and system-level repositories to install dependencies and NLP models. These network operations are strictly limited to trusted or well-known services and are essential for the skill's primary librarian purpose.\n- [DATA_EXFILTRATION]: No evidence of sensitive data exposure or exfiltration was found. The skill implements path traversal protection in scripts/utils/script_utils.py by validating that all base directory resolutions remain within the skill's own directory or the repository root.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted documentation content from the internet, which represents a potential attack surface.\n
Ingestion points: Documentation is fetched via the requests library from geminicli.com in scrape_docs.py.\n
Boundary markers: The skill organizes content into structured index entries and extracted subsections before being presented to the AI agent.\n
Capability inventory: The skill is granted Bash tool permissions and uses file system operations (write_file, replace).\n
Sanitization: The scraper uses markdownify and BeautifulSoup to process HTML content into Markdown, providing a layer of sanitization for the ingested data.\n- [DYNAMIC_EXECUTION]: Standard dynamic loading patterns are used for handling optional dependencies, such as importlib.import_module for the yaml library. The skill also processes metadata provided as JSON via command-line arguments in scripts/management/manage_index.py, which is a standard operational pattern for this librarian tool.

gemini-cli-docs