The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes the google-genai Python package and the gemini CLI. Both resources are provided by Google, a well-known and trusted service provider. The installation instructions point to official and reputable sources.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest user-provided research topics, outlines, and local data directories, as well as fetch information from the public internet. This processing of external, untrusted content constitutes a surface for indirect prompt injection.
Ingestion points: Local files are ingested via the data_dir and attachments arguments in scripts/deep_research.py and scripts/lite_research.sh.
Boundary markers: The prompt construction in build_research_prompt and scripts/lite_research.sh uses markdown headers to separate sections, but does not implement strict sanitization or specific escape sequences for external data.
Capability inventory: The skill has the capability to read local files, write reports to the filesystem, and perform network operations via the Google GenAI API.
Sanitization: There is no explicit sanitization of the input topic or outline before they are interpolated into the system prompts. However, the skill's documentation in docs/guide.md explicitly warns the user about risks like prompt injection via files and web content exfiltration.

gemini-deepresearch