The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to generate .tape files and execute them using the vhs command-line tool. Because .tape files contain shell-like instructions that the tool executes in a terminal environment, this process involves the creation and execution of dynamically generated code.
[PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection because it transforms user-supplied CLI flows into executable scripts. A malicious user could provide input designed to trick the agent into embedding harmful shell commands within the generated tape.
Ingestion points: User prompts describing the CLI commands and terminal outputs to be recorded in the demo.
Boundary markers: The skill instructions explicitly warn the agent and the user to treat the generated tape as executable code and review it before rendering.
Capability inventory: The vhs tool executes commands in a terminal session, granting the generated script access to any binary, file, or environment variable available to the user's shell.
Sanitization: The instructions recommend using fake credentials and reviewing all commands for safety before execution.
[EXTERNAL_DOWNLOADS]: The skill references external resources including the vhs tool, its dependencies (ttyd, ffmpeg), and the charmbracelet/vhs-action for GitHub Actions. These are recognized as well-known developer tools from an established vendor.

tape-demo