A new agentic AI use case on energy sector subsurface simulation

[Zenodia]

open source energy sector use case for subsurface simulation workload, there is an orchestrator agent which can trigger 2 workloads :

• Simulator Agent : Conversational assistant: keywords, parsing, running cases, plotting, flow diagnostics.
• Workflow Agent: Agentic optimization of simulation-intensive workflows: optimization, history matching, uncertainty quantification.

Implemented Features :

Pattern	Location	Description	Benefit
Query decomposition	sim_agent/.../query_decomposition.py, TOOL_DECISION_TREE.md	Structured routing via a decision tree; the LLM outputs JSON plans rather than relying solely on ReAct. Includes fallback heuristics when JSON parsing fails.	Single LLM call produces a plan; tools run deterministically—fewer calls than ReAct, faster response, lower cost, and better fit for smaller/cheaper models.
Agent skills	sim_agent/.../skills/*/	Modular skills with SKILL.md, tool auto-discovery, and skill_loader. Each skill is self-contained (scripts, references, assets).	Plug-and-play: add or remove skills without touching the agent core; easier to maintain, reuse, and share across projects.
External tools as skills	sim_agent/.../skills/results_skill/	Example of wrapping an external library (pyflowdiagnostics) as an agent tool. Add your own skills with your own tools by following the same structure.	Integrate any library or CLI without changing the agent.
Human-in-the-loop (HITL)	query_decomposition.py, runner.py	Optional confirmation before modify/run/plot; configurable via confirm_before_modify, confirm_before_run, etc.	Safety and control for sensitive actions (e.g., modifying files, running simulations). Important for production and auditability.

Please follow the README.md for environment build as well as run through different example queries to trigger different workloads.