The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface where untrusted data is processed to influence agent behavior and tool calls.\n
Ingestion points: Processes external specification files provided via seed_path and user signals delivered during runtime via ouroboros_session_signal (SKILL.md).\n
Boundary markers: Employs a structured job_observer contract with explicit verification steps to ensure data integrity during agent-to-agent handoffs.\n
Capability inventory: Utilizes tools for background job execution (ouroboros_start_execute_seed), session management, and the orchestrate_subagents capability for monitoring.\n
Sanitization: Includes 'fail-closed' instructions for validating the structural integrity of session metadata and contracts.\n- [DYNAMIC_EXECUTION]: The skill utilizes the orchestrate_subagents capability to spawn child agents that monitor background tasks.\n
Evidence: Instructions mandate spawning an observer by passing a dynamically recovered job_observer contract (SKILL.md).\n
Risk: While the observer is restricted to read-only operations, the dynamic instantiation of sub-agents based on session metadata represents a complex execution flow.

ouroboros-run