The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill executes a local Python script (session_manager.py) to update the session state during the spawn process.
[COMMAND_EXECUTION]: The skill constructs a shell command to update session state: python {skill_path}/scripts/session_manager.py --update {session-id} --state running. The {session-id} variable is provided by the user via the command line (e.g., /hub:spawn 20260317-143022). If this input is not sanitized by the execution environment, it could potentially allow for argument injection or command chaining.
[INDIRECT_PROMPT_INJECTION]: The skill uses templates and configuration files to build instructions for parallel subagents, creating a surface for indirect injection.
Ingestion points: Data is read from .agenthub/sessions/{session-id}/config.yaml and references/agent-templates.md (specifically the {task} and {variables} fields).
Boundary markers: The prompt includes a 'Constraints' section explicitly instructing subagents not to modify or read other agents' work, which serves as a natural language boundary.
Capability inventory: Subagents are granted the capability to modify files, run tests, and commit changes within their isolated worktrees.
Sanitization: The instructions do not specify any programmatic sanitization or escaping of the variables interpolated into the Agent() tool prompts.

spawn