The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the tmux CLI to orchestrate terminal windows and panes. This involves spawning new processes (claude-kimi) and sending simulated keystrokes to them.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It reads content from an external terminal pane using tmux capture-pane and integrates this data into its own reasoning context. * Ingestion points: Data is ingested from the teammate's tmux pane via capture-pane in SKILL.md. * Boundary markers: No specific delimiters or 'ignore' instructions are used to wrap the captured external content. * Capability inventory: The skill executes various shell commands via tmux and manages files in /tmp. * Sanitization: The captured terminal output is not sanitized or escaped before being processed by the agent.
[COMMAND_EXECUTION]: Potential for command injection via unsanitized task identifiers. The skill constructs shell commands and file paths using a 'taskslug' derived from the current task. If an attacker can influence the task description to include shell metacharacters (e.g., ;, &, |) or path traversal sequences, it could lead to unauthorized command execution or file manipulation.

claude-kimi-teammate