The Agent Skills Directory

[PERSISTENCE]: The installer script (scripts/install.sh) modifies the user's ~/.zshrc file to add a shell hook. This hook (assets/agent-shell-hook.zsh) automatically executes the agent-observe script every time a command is entered or finishes in a tmux pane, ensuring the cockpit's tracking logic persists across new shell sessions.\n- [COMMAND_EXECUTION]: The skill makes extensive use of the tmux CLI and Python's subprocess module to manage terminal panes, send keys, and execute arbitrary commands. Specifically, scripts/agent-pane and scripts/agent-run are designed to execute user or agent-supplied commands within tmux windows.\n- [INDIRECT_PROMPT_INJECTION]: The status classification engine in scripts/agent_classify.py parses terminal output captured via tmux capture-pane. A malicious process or data could output specific strings (e.g., AGENT_STATUS: done) to manipulate the monitoring agent's state classification.\n
Ingestion points: Terminal output captured in scripts/agent-status-scan and scripts/agent-read (via tmux capture-pane).\n
Boundary markers: None; the classifier searches for regex patterns within the raw terminal buffer tail.\n
Capability inventory: The skill can launch new processes, send keystrokes to active panes, and manipulate git worktrees.\n
Sanitization: Terminal output is not sanitized before being evaluated against status detection patterns.\n- [DYNAMIC_EXECUTION]: The skill dynamically assembles shell command wrappers in scripts/agent-pane and generates tmux configuration files during installation using sed to substitute absolute paths into the generated configuration.\n- [EXTERNAL_DOWNLOADS]: The skill configuration in assets/tmux-agent-persistence.conf references external tmux plugins (tmux-resurrect, tmux-continuum) and the installer script provides instructions to download the Tmux Plugin Manager (TPM) from its official GitHub repository.

tmux-agent-herdr-lite