Catgo Build Workflow

catgo-build-workflow

Tight playbook for assembling a CatGo workflow in one or two MCP round-trips. The default CatBot path explores node_types, node_details, templates, then incrementally adds nodes and edges — that is 8+ MCP calls and the user sees the "Thinking…" indicator for tens of seconds. This skill cuts it to 1–2 calls.

Iron rule: every reaction-mechanism workflow ends in Gibbs free energy

Activities, overpotentials, and barriers reported in catalysis literature are differences in Gibbs free energy at the operating temperature, not DFT electronic energies. So whenever the user asks for a reaction pathway — CO2RR, OER, HER, NRR, ORR, NEB, slow-growth, C–N coupling, anything ending in *RR, anything called "free energy diagram" or "volcano plot" — the workflow must contain a freq node between geo_opt (or md) and free_energy (or the reaction-specific analysis node). Without freq there is no ZPE and no thermal/entropic correction, the resulting numbers cannot be compared to experiment, and the user is silently wrong.

If the user proposes a mechanism workflow without a freq step, add one anyway and tell them one short sentence why ("Inserted a freq step so the ΔG values include ZPE + TS — without it the free-energy diagram is just an electronic-energy diagram"). If they explicitly say "skip freq for now, I just want a quick electronic-energy scan", honour it but flag that the result is not a Gibbs energy.

The freq node must run on the same geometry as the final relaxation it sits after — chaining geo_opt → freq → free_energy keeps the geometries consistent. For adsorbate thermochemistry, use freeze_mode: "adsorbate": freeze the full slab and vibrate only atoms tagged by adsorbate_place.

Clean-slab invariant

Every surface-reaction workflow built from a bulk structure must use structure_input → slab_gen → geo_opt(clean slab) → adsorbate_place. Reuse the same relaxed clean-slab output for every adsorbate branch. Never connect slab_gen directly to adsorbate_place: doing so optimizes only slab+adsorbate states and leaves no consistent relaxed clean-slab reference.

When to use this skill

• User asks to create, build, set up, or make a workflow.
• User names a known reaction or pipeline (CO2RR, OER, HER, NRR, NEB, DOS, slow-growth, bulk→slab→adsorbate).
• User pastes a textual recipe like "structure_input → geo_opt → freq → free_energy".

If the user wants to modify an existing workflow (add a node to one that already exists), prefer a direct catgo_workflow {action:"add_node"} call rather than reloading this skill.

The fast path

1. For stock HER/OER/ORR/NRR/CO2RR workflows, call catgo_quickbuild and stop; its server-side recipes enforce the clean-slab invariant. Use the manual path below only for custom graphs.
2. Pick a recipe from the "Recipes" section below — or assemble one from the node-type table — and prepare the graph_json payload.
3. Call catgo_workflow with action="create", name="<descriptive>", template_id only if you genuinely want the backend's stock template (most of the time you do not, because the recipes here are tighter). Otherwise omit template_id — create will auto-add a structure_input node seeded from the viewer's current structure.
4. Immediately call catgo_workflow with action="batch" and an operations array carrying every add_node + connect step in one round-trip. Do not call add_node one at a time.
5. Confirm with one short sentence ("Built '<name>': N nodes, M edges. Open the Workflow tab to inspect."). Do not list every node — the user can see the graph in the editor.

That is the entire happy path. Do not call templates, node_types, node_details, list_presets, or get before creating unless the user explicitly asks "what templates exist?" — those calls only exist for discovery and the recipes below already cover the common cases.

Recipes

Each recipe gives the operations array you pass to batch. The seed structure_input node is already created for you by create; reference it as "si" in from fields. Use stable short IDs (n1, n2, …) for new nodes — these only need to be unique within the workflow.

CO2RR (CO2 reduction on metal slab)

[
  {"op":"add_node","id":"opt","type":"geo_opt","x":300,"y":200,"params":{"software":"vasp","encut":520,"ediffg":-0.03}},
  {"op":"add_node","id":"freq","type":"freq","x":520,"y":200,"params":{"software":"vasp","freeze_mode":"adsorbate"}},
  {"op":"add_node","id":"fe","type":"free_energy","x":740,"y":200,"params":{"temperature":298.15,"reference":"CHE"}},
  {"op":"connect","from":"si","to":"opt"},
  {"op":"connect","from":"opt","to":"freq"},
  {"op":"connect","from":"freq","to":"fe"}
]

For multi-intermediate CO2RR (CO2* → COOH* → CO* → CHO* …), duplicate geo_opt + freq per intermediate, all wired to the same free_energy node which aggregates ΔG values.