haipipe-nn-modelset

name	haipipe-nn-modelset
description	Layer 4 (ModelSet / Pipeline) specialist of haipipe-nn: composes multiple ModelInstances into a registry-backed pipeline. Called by /haipipe-nn; direct invocation works layer-scoped.
argument-hint	[function] [args...]
allowed-tools	Bash, Read, Write, Edit, Grep, Glob
metadata	{"version":"0.1.1","last_updated":"2026-07-04","summary":"Layer 4 (ModelSet / Pipeline) specialist of haipipe-nn."}

Skill: haipipe-nn-modelset

Layer 4 specialist. Owns ModelSet (pipeline) composition — combining multiple ModelInstances into a registry-backed serving pipeline (e.g. an ensemble, a router, a multi-stage forecast). Called by the /haipipe-nn orchestrator; can also be invoked directly.

Function axis: dashboard | review | generate | test

---

Commands

/haipipe-nn-modelset                      -> dashboard: registered ModelSets
/haipipe-nn-modelset dashboard            -> same
/haipipe-nn-modelset review [set_name]    -> structural review of a ModelSet
/haipipe-nn-modelset generate             -> compose a new ModelSet from instances
/haipipe-nn-modelset test [set_name]      -> smoke-test the ModelSet end-to-end

---

Dispatch Table

Invocation    This skill's ref       Umbrella's fn doc
------------- ---------------------- ------------------------------
dashboard     ref/concepts.md        ../haipipe-nn/fn/fn-dashboard.md
review        ref/concepts.md        ../haipipe-nn/fn/fn-review.md
generate      ref/concepts.md        ../haipipe-nn/fn/fn-generate.md
test          ref/concepts.md        ../haipipe-nn/fn/fn-test.md
(no fn arg)   ref/concepts.md        (ref-only mode)

L4 is the terminal layer of haipipe-nn — generate does NOT need a downstream layer ref. The next consumer is /haipipe-end (Stage 6).

---

Step-by-Step Protocol

Step 0: Read ../haipipe-nn/ref/overview.md. Mandatory. Step 1: Parse args. Function vocabulary: dashboard | review | generate | test. Step 2: Read this skill's ref/concepts.md for L4 specifics. Step 3: Read the umbrella fn doc. Step 4: Execute, scoped to L4 (composition only — no instance retraining). Step 5: Emit the structured tail.

---

Layer Scope

Owns:

• ModelSet definitions in code/haipipe/model_base/ (modelinstance_set.py, modelinstance_pipeline.py)
• Registry of which instances participate in which set
• Cross-instance routing / ensemble logic

Upstream dependency (L3): Reads _WorkSpace/5-ModelInstanceStore/{modelinstance_name}/{version}/ directories. A ModelSet is a referenced registry; if an instance is missing, escalate to /haipipe-nn-instance review.

Hand-off contract (L4 -> Stage 6): ModelSet name + member ModelInstance list is what /haipipe-end-endpointset package consumes. Layout must satisfy the Endpoint_Pipeline's expectations.