The Agent Skills Directory

[PERSISTENCE]: The skill documents and provides verification steps for iTerm2's AutoLaunch functionality. This feature allows scripts placed in ~/Library/Application Support/iTerm2/Scripts/AutoLaunch/ to execute automatically upon application startup, providing a mechanism for persistent code execution on the host system.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which could influence agent behavior or lead to command execution.
Ingestion points: Workspace configuration is read from and written to ~/.config/iterm2/layout.toml by the agent and external scripts.
Boundary markers: The instructions lack specific delimiters or warnings to ignore embedded instructions within the TOML configuration file.
Capability inventory: The skill uses Bash, Read, and Edit tools. The system it configures executes shell commands (such as those defined in the [commands] section of the TOML) within active terminal panes.
Sanitization: No sanitization or validation logic is defined for the command strings ingested from the TOML configuration.
[PROMPT_INJECTION]: The "Self-Evolving Skill" section contains meta-instructions directing the agent to autonomously update the SKILL.md file. This self-modification capability could be exploited by an adversary to induce the agent to persist malicious instructions or bypass safety guidelines by manipulating the environment to trigger specific "workarounds" or "fixes."
[COMMAND_EXECUTION]: The skill uses a Bash HEREDOC script during setup to perform file system operations (mkdir, cp). Additionally, the troubleshooting section suggests executing Python one-liners for configuration validation.

iterm2-layout