The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The instructions direct users to download code from https://github.com/https-deeplearning-ai/agentic-ai-public.git. The organization name 'https-deeplearning-ai' is highly suspicious as it appears to mimic 'deeplearning-ai', a well-known AI education platform. This pattern of using 'https-' as a prefix in an organization name is often used in typosquatting or impersonation attacks to gain user trust while delivering untrusted content.\n- [REMOTE_CODE_EXECUTION]: By instructing the user to build and run a Docker container based on the source code downloaded from the suspicious repository, the skill effectively facilitates the execution of remote code from an unverified source on the user's system.\n- [COMMAND_EXECUTION]: The skill provides and encourages the use of shell commands (git clone, docker build, docker run) to fetch and execute the external code. This creates a direct path for the execution of potentially malicious scripts or binaries included in the suspicious repository.\n- [PROMPT_INJECTION]: The system is designed to ingest data from external sources such as Tavily, arXiv, and Wikipedia. This creates a surface for indirect prompt injection, where an attacker could place malicious instructions in web pages or academic abstracts to manipulate the agent's behavior when it processes those sources.\n
Ingestion points: Web and academic search tools defined in src/research_tools.py.\n
Boundary markers: None identified in the provided snippets to separate untrusted tool output from system instructions.\n
Capability inventory: The system operates within a Docker container, has network access for API calls, and maintains state in a local PostgreSQL database.\n
Sanitization: The documentation does not specify any sanitization or validation of the content retrieved from external tools before it is processed by the LLM agents.

agentic-research-agent-fastapi