The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill clones the project source code from a repository belonging to a recognized educational organization in the AI field.
[INDIRECT_PROMPT_INJECTION]: The system is designed to ingest and summarize information from external sources (Tavily, arXiv, Wikipedia), which presents an attack surface for instructions potentially embedded in third-party content.
Ingestion points: Data retrieval functions in src/research_tools.py fetching web results and research paper summaries.
Boundary markers: The LLM prompts provide task definitions but lack specific delimiters or instructions to disregard malicious commands embedded in the retrieved source materials.
Capability inventory: The agent has the ability to modify local database state via SQLAlchemy and perform outbound network requests to search APIs.
Sanitization: External content is synthesized directly by the LLM for report generation without prior automated sanitization or filtering.
[COMMAND_EXECUTION]: The documentation includes instructions for users to manually execute shell commands for environment setup, repository cloning, and Docker container orchestration.
[CREDENTIALS_UNSAFE]: The application code provides a hardcoded default connection string (postgresql://app:local@127.0.0.1:5432/appdb) used for local development and database state management.

agentic-research-agent-deeplearning