The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's core functionality is to execute arbitrary shell commands on the host system to capture their output. While intended for generating demos, this capability allows for general command execution which can be risky if the agent processes untrusted instructions.
Evidence: scripts/auto_generate_demo.py and scripts/record_interactive.sh both invoke the vhs tool to run and record shell commands.
[DYNAMIC_EXECUTION]: The skill dynamically constructs and executes .tape files, which are scripts for the vhs recording tool. This dynamic generation is the mechanism used to execute the user-provided commands.
Evidence: scripts/auto_generate_demo.py writes command strings directly into a temporary .tape file before calling the vhs binary.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to injection attacks because it fails to sanitize command strings before inserting them into .tape files. Specifically, scripts/auto_generate_demo.py does not escape double quotes, which could allow a malicious command to terminate the Type directive and inject other vhs directives. This could be used to redirect the tool's output and overwrite or corrupt files.
Ingestion points: The --command arguments in scripts/auto_generate_demo.py and the commands field in configuration files used by scripts/batch_generate.py.
Boundary markers: None. Command strings are directly interpolated into the file template.
Capability inventory: Execution of commands in a terminal shell and control over tool output paths.
Sanitization: None. The script uses f-string interpolation: f'Type "{cmd}" Sleep 500ms'.

cli-demo-generator