The Agent Skills Directory

[COMMAND_EXECUTION]: The skill enables the agent to execute arbitrary shell commands in background terminal sessions using tmux send-keys. This functionality is the primary purpose of the skill, intended for managing long-running tasks such as development servers, continuous builds, and remote SSH connections that exist outside the agent's direct execution context.
[PROMPT_INJECTION]: The skill introduces an indirect prompt injection surface by reading external process output into the agent's context through tmux capture-pane (found in SKILL.md). Ingested data from sources like server logs or remote shells could theoretically contain instructions intended to influence the agent. Following the mandatory evidence chain: 1) Ingestion points: terminal output captured via capture-pane; 2) Boundary markers: absent in code but safety warnings are provided in prose; 3) Capability inventory: full shell access via send-keys; 4) Sanitization: the skill recommends manual cleanup of ANSI escape sequences using sed. The risk is further mitigated by explicit documentation advising the agent not to automatically respond to security-sensitive prompts like sudo passwords or SSH host key verifications.

tmux