The Agent Skills Directory

[SAFE]: No malicious patterns or critical security vulnerabilities were detected.
[DATA_EXFILTRATION]: The skill transfers local files and user queries to the Google Gemini API. This interaction with a well-known service is the core functionality of the research tool and is documented accordingly.
[EXTERNAL_DOWNLOADS]: External Python dependencies such as google-genai and rich are managed through uv run and retrieved from official registries.
[PROMPT_INJECTION]: The skill processes external data (local files and previous research), creating a surface for indirect prompt injection.
Ingestion points: Processes local project files via context stores and previous research results via the --follow-up flag in scripts/research.py.
Boundary markers: Uses <previous_findings> tags to separate historical data from the current query instructions.
Capability inventory: Includes file system access (read/write for reports and state) and network operations to the Google Gemini API.
Sanitization: Includes logic in scripts/research.py to escape backticks and remove injection-prone tags from historical research data to prevent hijacking of the new research session.

deep-research