The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions direct the user to add a plugin from a repository located at https://github.com/jarrodwatts/claude-hud. This repository is owned by an individual user rather than a recognized trusted organization.
[REMOTE_CODE_EXECUTION]: The process of adding a marketplace and installing a plugin (/plugin marketplace add, /plugin install) involves fetching external code from a remote source and executing it as part of the agent's statusline HUD.
[COMMAND_EXECUTION]: The skill provides commands that modify the environment and configuration of the agent, including /claude-hud:setup and manual adjustments to environment variables (e.g., TMPDIR=~/.cache/tmp) for troubleshooting.
[INDIRECT_PROMPT_INJECTION]: The HUD is designed to monitor and display tool activity, agent status, and context usage. This functionality creates an attack surface where instructions embedded in processed data (such as tool outputs or task descriptions) could potentially be rendered in the HUD, posing a risk if the data is not properly sanitized.
Ingestion points: The HUD displays real-time tool activities, agent status, and todo progress derived from the current session context (SKILL.md).
Boundary markers: No boundary markers or "ignore embedded instructions" warnings are documented for the data displayed in the statusline.
Capability inventory: The skill requires environment-modifying commands and the execution of a persistent UI plugin (SKILL.md).
Sanitization: There is no evidence in the instructions that the plugin performs sanitization on the metadata it monitors.

claude-hud