The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external websites to perform market research, which exposes the agent to potential indirect prompt injection attacks.
Ingestion points: Data enters the context through tavily/search-assistant, infsh/agent-browser, and tavily/extract which fetch content from arbitrary URLs provided during research.
Boundary markers: None identified; instructions do not specify using delimiters or explicit warnings to the agent to isolate untrusted web content.
Capability inventory: The skill uses the belt CLI to execute search, browsing, Python code execution (infsh/python-executor), and image processing.
Sanitization: No sanitization or validation of the ingested external content is performed before it is processed by the agent.
[COMMAND_EXECUTION]: The skill interacts with the system environment using the belt CLI, invoking various application sub-commands via the Bash tool to perform automated research tasks.
[DYNAMIC_EXECUTION]: The skill includes a Python script template to be executed by the infsh/python-executor tool. This script dynamically generates a positioning map using the matplotlib library at runtime.
[EXTERNAL_DOWNLOADS]: The skill documentation instructs the user to install external software (belt-sh/cli) and references installation guidelines hosted on a remote GitHub repository (github.com/inference-sh).

competitor-teardown