The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes heavy use of the tmux CLI to create windows, send keys, and manage processes. It executes shell commands like git, command -v, and tmux capture-pane to manage the lifecycle of tickets. Arbitrary commands defined in the verify field of the manifest or inferred from project files are executed to validate ticket completion.
[REMOTE_CODE_EXECUTION]: The skill identifies and invokes other AI agents (codebuddy, claude) using the --permission-mode bypassPermissions flag. This explicitly disables security guardrails for the worker agents, allowing them to execute code and modify files without user confirmation. The skill passes user-provided or document-inferred prompts directly into these bypassed environments.
[DATA_EXFILTRATION]: While the skill primarily targets local task automation, it has the capability to read files via git show and tmux capture-pane. The automated invocation of external CLI agents (claude, codebuddy) which have network access, combined with bypassPermissions, creates a potential path for data to be sent to external endpoints if a ticket's prompt or the inferred project context is malicious.
[PRIVILEGE_ESCALATION]: The instruction to use --permission-mode bypassPermissions when spawning worker agents is a form of privilege escalation within the AI agent's environment, removing human-in-the-loop safety requirements for all sub-tasks.

implement-tmux