nmdc-metadata-suggestor

name	nmdc-metadata-suggestor
description	Use this skill to orchestrate NMDC metadata suggestions — detects input shape, routes to env-triad and/or metadata-suggestion-pipeline, and returns a merged LLMOutput JSON. Entry point for all metadata suggestion requests.

NMDC Metadata Suggestor Orchestration Skill

Receives data, determines what to run, returns LLMOutput JSON. No user interaction.

---

Step 1 — Detect input shape

Condition	Shape
Input has top-level key metadata_submission	Submission portal object
Input is a list of dicts with biosample-like keys (id, lat_lon, env fields)	ETL samples
Input is a list of dicts with MIxS column-name keys	Submission portal-shaped samples

---

Step 2 — Detect requested output

Condition	Route to
Input explicitly names env_triad, env_broad_scale, env_local_scale, or env_medium	env-triad only
Input is samples only (no submission object)	env-triad only
Input is a submission object with no sample list	metadata-suggestion-pipeline only
Input is a submission object that includes samples	Both — run metadata-suggestion-pipeline first, then env-triad

---

Step 3 — Execute and return

Run the routed skill(s) and return a single LLMOutput JSON object.

If both pipelines run, merge their metadata_fields lists into one response.

IMPORTANT: Do NOT call StructuredOutput until all routed skills have finished executing. Call StructuredOutput exactly once at the very end with the complete merged LLMOutput JSON.