The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script (session_manager.py) to manage the lifecycle and status of agent sessions. This is a standard administrative function for this skill type.
[INDIRECT_PROMPT_INJECTION]: The skill interpolates session data and template content into subagent prompts, creating a potential surface for indirect instructions.
Ingestion points: Data is loaded from .agenthub/sessions/{session-id}/config.yaml and template files in ../agenthub/references/agent-templates.md.
Boundary markers: The dispatch prompt uses basic variable substitution (e.g., {task}, {variables}) without explicit instruction boundary delimiters.
Capability inventory: The skill has the ability to spawn new agent instances via the Agent tool and perform filesystem writes to the .agenthub/board/ directory.
Sanitization: The skill lacks explicit sanitization for interpolated session variables, though it enforces strict filesystem isolation (git worktrees) for the resulting agents.
[SAFE]: The skill enforces security best practices for parallel agent execution, specifically requiring mandatory isolation: "worktree" to prevent cross-agent data contamination and filesystem interference.

spawn