The Agent Skills Directory

[COMMAND_EXECUTION]: The skill leverages the sesh command-line utility and TUI to manage tmux sessions, daemon processes, and multi-machine networking. This includes lifecycle management tasks like starting, stopping, and restarting services via supervisorctl.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data via multiple ingestion points: ticket create --prompt (tickets-and-blobs.md), sesh thread capture (threads.md), and sesh blob add (tickets-and-blobs.md). Boundary markers like 'guarded bracketed paste' and 'literal RPC steering' (delivery.md) are used to prevent malicious payloads from being interpreted as commands. The tool's capabilities include subprocess execution via the sesh CLI, file system reads/writes via blob commands, and mesh-wide network operations (fleet.md). Sanitization is implemented via ARGV-based substitution for plugins (integrations.md) and metadata prepending for ticket prompts.
[PRIVILEGE_ESCALATION]: Instructions include management of system-level services using supervisorctl. While this typically involves elevated permissions, the skill frames it as the authorized procedure for maintaining the tool's daemon in production environments.

sesh-cli