The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local script scripts/validate_orchestration.py to verify design completion and complexity gates. The script uses standard JSON parsing and lacks dangerous primitives such as eval(), exec(), or network operations.
[PROMPT_INJECTION]: The orchestrator is designed to ingest and merge state_patch updates from external capabilities into its primary state, creating an indirect prompt injection surface.
Ingestion points: Data enters the context via state_patch objects defined in SKILL.md and references/handoff-v1.md.
Boundary markers: The instructions do not define specific delimiters (e.g., XML tags) or warnings to the agent to ignore instructions embedded within the state data.
Capability inventory: The skill possesses the ability to execute the scripts/validate_orchestration.py subprocess.
Sanitization: There is no evidence of validation or sanitization of the content within the state_patch before it is merged into the system_design_state.

system-design-orchestrator