The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a workflow where data retrieved from external tools (Tavily, arXiv, and Wikipedia) is directly interpolated into prompts for the writer and editor agents. This creates a surface where malicious content embedded in research sources could potentially influence the behavior of the agent during the report generation phase.
Ingestion points: User-provided research prompts in main.py and external search results from tavily_search_tool, arxiv_search_tool, and wikipedia_search_tool in src/research_tools.py.
Boundary markers: Absent; search results are interpolated directly into f-strings used for LLM completions.
Capability inventory: Writing results to a Postgres database, making external API calls to search engines and LLM providers.
Sanitization: No sanitization or filtering of external tool output is performed before it is sent to the LLM.
[CREDENTIALS_UNSAFE]: The skill contains hardcoded default credentials for a local PostgreSQL database (postgresql://app:local@127.0.0.1:5432/appdb) in the connection string and environment variable examples. While appropriate for local development and Docker-based tutorials, these should be replaced with secure secrets management for production environments.

agentic-ai-research-agent