The Agent Skills Directory

[SAFE]: The skill incorporates a robust safety framework. The safety-check.sh script monitors for dangerous command patterns (e.g., rm -rf, mkfs) and sensitive data patterns in payloads before they are sent to terminal panes. Most scripts support a --dry-run flag to allow for safe previewing of actions.
[DATA_EXFILTRATION]: Scripts like auto-tmux.sh and remote-readonly.sh capture terminal output but include a redact function that automatically masks sensitive information such as OpenAI/Anthropic API keys, GitHub tokens, and Bearer tokens using regular expression filtering.
[COMMAND_EXECUTION]: The core functionality relies on executing tmux commands and local shell scripts. The skill follows the principle of least privilege by providing a remote-readonly.sh script specifically for remote diagnostics that is restricted to non-mutating tmux commands via SSH.
[INDIRECT_PROMPT_INJECTION]: As the skill reads arbitrary terminal output (e.g., via capture-pane), it possesses an attack surface for indirect prompt injection. However, the risk is mitigated by the skill's architecture which defines specific roles (Commander, Worker) and structured reporting protocols.
[SAFE]: The skill's configuration and assets are managed via Git submodules and relative soft links to trusted or internal paths, avoiding the use of unverifiable external downloads during runtime.

auto-tmux