The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from session configurations (config.yaml) and task definitions, which are interpolated into the system prompts of spawned sub-agents. This creates a surface where instructions embedded in session data could influence sub-agent behavior.
Ingestion points: Processes data from .agenthub/sessions/{session-id}/config.yaml and task descriptions provided during the spawning process.
Boundary markers: The skill does not explicitly use delimiters or guardrail instructions to isolate interpolated task data from the sub-agent's core instructions.
Capability inventory: The skill utilizes the Agent tool to spawn new sub-agents and executes a local Python script for session state management.
Sanitization: No explicit sanitization or validation of the variables extracted from the session config is performed before they are placed into the prompt template.
[DYNAMIC_EXECUTION]: The skill executes a local Python script (session_manager.py) located within the skill's own directory to update session status. This is a standard management function for an orchestration skill.
Evidence: Call to python {skill_path}/scripts/session_manager.py --update {session-id} --state running.

spawn