The Agent Skills Directory

[DYNAMIC_EXECUTION]: The calculate tool function implementation in SKILL.md uses the Python eval() function to process the expression argument. Since this argument is generated by an LLM based on user input, it allows for arbitrary code execution on the host system where the agent is running.
[COMMAND_EXECUTION]: The skill provides an example of LocalCommandLineCodeExecutor in SKILL.md. This component allows an agent to execute arbitrary shell commands on the local system. While the skill correctly suggests using Docker as a secure alternative, the inclusion of local execution represents a high-privilege capability that could be abused if the agent is compromised.
[INDIRECT_PROMPT_INJECTION]: The skill describes a multi-agent environment where agents ingest data from various sources (e.g., the researcher agent).
Ingestion points: The researcher and tool_agent process task descriptions and potentially external information.
Boundary markers: None are implemented in the code snippets provided.
Capability inventory: Includes tool execution (with eval) and shell command execution via LocalCommandLineCodeExecutor.
Sanitization: There is no evidence of input validation or sanitization before passing data to the eval() function or the code executor.

autogen-development