The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources (Tavily, arXiv, Wikipedia) and interpolates it directly into prompts for the writer and editor agents without using boundary markers or sanitization.
Ingestion Points: Data enters through tavily_search_tool, arxiv_search_tool, and wikipedia_search_tool in src/research_tools.py.
Boundary Markers: The skill does not employ delimiters or specific instructions to ignore embedded commands within the retrieved research data.
Capability Inventory: The agent has capabilities for web search, academic database querying, and multi-step task execution.
Sanitization: There is no evidence of filtering or escaping logic applied to the external content before it is processed by the LLM.
[COMMAND_EXECUTION]: The documentation provides instructions for executing shell commands and managing services, such as Docker container execution, database cluster management via pg_ctlcluster, and SQL querying via psql.
[CREDENTIALS_UNSAFE]: The skill includes a hardcoded default database connection string (postgresql://app:local@127.0.0.1:5432/appdb) in main.py. While intended for local development environments, it follows the pattern of hardcoding credentials.

deeplearning-ai-agentic-research