The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill monitors terminal output from multiple sub-agents to orchestrate tasks. This creates a surface for indirect prompt injection if a sub-agent processes malicious data and outputs text that influences the supervisor's logic.
Ingestion points: The scripts classify-pane.sh, poll-cycle.sh, and run-loop.sh capture sub-agent pane output via tmux capture-pane.
Boundary markers: While the skill looks for specific markers like ORCHESTRATOR:DONE, there are no delimiters or instructions used to isolate untrusted sub-agent output from the supervisor's primary instructions.
Capability inventory: The skill has the ability to execute commands via tmux send-keys, perform Git operations, use the GitHub CLI (gh), and modify local state files.
Sanitization: The skill strips ANSI color codes but does not sanitize the text content for potential prompt injection patterns before it is processed by the AI.
[COMMAND_EXECUTION]: The skill utilizes tmux send-keys to programmatically inject commands and inputs (such as approval signals like y and Enter) into other shell sessions. This allows the agent to control and execute actions in multiple terminal environments.
[DYNAMIC_EXECUTION]: The spawn-agent.sh script employs python3 -c to generate a UUID for session tracking. This represents a simple use of dynamic execution for a utility task using a static template.

orchestrate