The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use the belt CLI tool for various operations, including authentication (belt login) and executing remote applications via belt app run. This is the standard operational mode for the associated platform.
[EXTERNAL_DOWNLOADS]: The documentation references installation instructions hosted on GitHub (raw.githubusercontent.com/inference-sh/skills/refs/heads/main/cli-install.md). These resources are directly related to the vendor's own infrastructure and tools.
[REMOTE_CODE_EXECUTION]: The skill generates and executes Python code snippets via the infsh/python-executor application. The provided code is specifically designed for rendering charts using the matplotlib library and does not perform any suspicious system or network operations.
[REMOTE_CODE_EXECUTION]: The skill uses infsh/html-to-image to render HTML/CSS into images. This is a dynamic execution of markup content provided in the instructions for visual styling.
[PROMPT_INJECTION]: A potential surface for indirect prompt injection exists because the skill encourages the agent to interpolate data into Python and HTML templates.
Ingestion points: The agent is expected to process pitch deck content (problem statements, traction metrics) which could be provided by users.
Boundary markers: There are no explicit boundary markers or instructions to sanitize or escape user-provided content before inserting it into the html or python blocks.
Capability inventory: The skill utilizes belt app run which supports Python execution and HTML rendering. Access to these capabilities is restricted to the tools allowed in the frontmatter (Bash).
Sanitization: No explicit sanitization or validation of the generated code/markup is described in the skill body.

pitch-deck-visuals