The Agent Skills Directory

[DYNAMIC_EXECUTION]: Documentation files references/tool-builder.md and references/agent-patterns.md provide code templates using the Python eval() function to process tool arguments (call.args['expression']). This pattern is inherently dangerous as it allows for arbitrary code execution if the input provided to the tool is not strictly validated.
[COMMAND_EXECUTION]: The skill configuration in SKILL.md allows the agent to execute any Python command using Bash(python *). While this level of access is functional for a developer SDK, it grants the agent significant capability to interact with the underlying system.
[EXTERNAL_DOWNLOADS]: The skill prompts for the installation of the inferencesh package from PyPI and mentions other external integrations like tavily/search-assistant and various AI models.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from AI models and user inputs which are interpolated into application logic, creating a surface for indirect injection attacks.
Ingestion points: Data processed via client.run() and agent.send_message() as shown in SKILL.md and references/agent-patterns.md.
Boundary markers: The provided examples do not demonstrate the use of delimiters or specific instructions to ignore embedded commands in ingested data.
Capability inventory: The skill is granted access to the Bash tool for running Python code, as specified in SKILL.md.
Sanitization: The code templates do not include sanitization or validation steps for external content before execution or processing.

python-sdk