The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The SKILL.md file provides instructions to download a Python script (starhtml_check.py) from an external GitHub repository (github.com/renatocaliari) using curl. It also suggests fetching demo files from github.com/banditburai.
[REMOTE_CODE_EXECUTION]: The skill directs users to download a remote script, place it in a system executable directory (/usr/local/bin/), and grant it execution permissions (chmod +x), facilitating the execution of remote code.
[COMMAND_EXECUTION]: The instructions involve executing system commands to install and update the starhtml-check utility, including commands that fetch and run code from the internet.
[DYNAMIC_EXECUTION]: The js() function documented in reference/js.md allows for the execution of raw JavaScript expressions. While the documentation provides a security warning, this capability allows for dynamic code execution at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill features a markdown rendering plugin (data_markdown) and direct JavaScript integration, creating an attack surface where untrusted data could influence application behavior.
Ingestion points: Untrusted data can enter through the todo_text parameter in SSE handlers or the data_markdown attribute.
Boundary markers: None identified in the provided documentation examples.
Capability inventory: Subprocess execution via starhtml-check (if run), raw JavaScript execution via js(), and network operations via get() and post() actions.
Sanitization: The documentation includes a warning regarding JavaScript injection but does not provide mandatory sanitization patterns for the markdown plugin.

cali-coding-starhtml