The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent and user to download and install software from external sources.
Fetches the project repository from https://github.com/hilash/cabinet.git using git clone.
Installs the @anthropic-ai/claude-code CLI globally via npm.
[REMOTE_CODE_EXECUTION]: The installation process utilizes npx create-cabinet@latest, which downloads and executes a bootstrapper package from the NPM registry at runtime.
[COMMAND_EXECUTION]: The skill documents a 'Web Terminal' feature that provides full shell access.
Powered by xterm.js and node-pty, this feature allows the execution of arbitrary shell commands within the browser.
The cabinet-daemon.ts automatically executes agent prompts through the claude CLI on a cron schedule.
[DYNAMIC_EXECUTION]: The skill features 'Embedded HTML Apps' that automatically render index.html files found in the data directory.
These files are rendered as iframes, allowing for the execution of arbitrary client-side JavaScript within the application context.
[INDIRECT_PROMPT_INJECTION]: The skill processes markdown and HTML files stored in the data/ directory, creating a surface for indirect instructions to be processed by agents.
Ingestion points: Markdown files (.md) and HTML files (.html) in the data/ directory are read and processed by the system (SKILL.md).
Boundary markers: None explicitly defined to separate data from instructions during agent processing.
Capability inventory: The system includes a background daemon that executes agent prompts, a web terminal with shell access, and file-writing capabilities (SKILL.md).
Sanitization: Not documented; the system executes shell commands via a terminal and renders local HTML files in iframes.

cabinet-ai-knowledge-base