The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources, which is a vector for indirect prompt injection. \n
Ingestion points: Untrusted data enters the agent context via infsh/agent-browser (screenshots and scraping) and tavily/extract (web content) as described in SKILL.md. \n
Boundary markers: The skill does not define delimiters or instructions to ignore potential commands embedded in the retrieved web content. \n
Capability inventory: The agent has permissions to execute arbitrary Python code via infsh/python-executor and run shell commands via the Bash(belt *) tool. \n
Sanitization: No sanitization or validation of the external content is present in the provided instructions. \n- [DYNAMIC_EXECUTION]: The skill utilizes a runtime environment to execute Python scripts for creating positioning maps. \n
It runs Python code that imports matplotlib to generate and save images. \n
While the script provided in SKILL.md uses a static template, this type of dynamic execution is a capability that requires monitoring. \n- [EXTERNAL_DOWNLOADS]: The skill suggests installing several external components for its core functionality. \n
It directs users to install belt-sh/cli and additional skills from the inference-sh GitHub repository. \n
These downloads are associated with the infrastructure provider for the skill. \n- [COMMAND_EXECUTION]: The skill relies on the belt CLI tool for all its major operations. \n
The Bash(belt *) permission enables the agent to execute research and analysis tasks via the belt command interface.

competitor-teardown