The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The documentation provides instructions to install the freeze CLI tool from a public GitHub repository. The suggested installation command includes GOSUMDB=off, which disables Go's checksum database verification; this is presented as a workaround for regional connectivity issues but bypasses an integrity check for the downloaded module.
[COMMAND_EXECUTION]: The skill executes local bash and Python scripts and leverages system binaries, specifically Google Chrome and the freeze utility, to perform rendering operations. The headless Chrome execution uses the --no-sandbox flag, which is a common requirement in certain environments but reduces browser process isolation.
[PROMPT_INJECTION]: The skill processes untrusted terminal output which is rendered into a visual format for the agent to interpret. This creates a surface for indirect prompt injection where text or formatting in the captured CLI stream could attempt to influence the agent's judgment through visual cues.
Ingestion points: Captured .ansi files processed by scripts/ansi2html.py and rendered by scripts/render_ansi.sh.
Boundary markers: No explicit delimiters or boundary markers are used to isolate the rendered content within the output image.
Capability inventory: The agent maintains its standard set of tools for file manipulation and command execution.
Sanitization: Input text is HTML-escaped during the conversion process, which prevents script injection into the intermediate HTML, though the visual content is preserved for the agent's vision system.

terminal-screenshot