The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes system commands like ps, kill, and tmux to perform its core functions. It incorporates defensive instructions to prevent command injection, such as advising against shell string concatenation and requiring re-validation of process IDs before termination.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes process logs and terminal history, creating a surface for indirect prompt injection from the output of external tasks.
Ingestion points: External data enters the context via tail, wc, and tmux capture-pane as described in SKILL.md.
Boundary markers: Explicit delimiters for process output are not defined in the instructions.
Capability inventory: The skill can terminate processes and execute user-defined fallback commands after obtaining confirmation.
Sanitization: Risk is mitigated by the requirement for independent confirmation (cross-referencing CPU and network signals) and human-in-the-loop validation for all high-impact actions.
[EXTERNAL_DOWNLOADS]: The documentation references an installation command that fetches components from the author's own repository (soia-team). This represents standard deployment of vendor-provided tools.

soia-dev-terminal-ops