haipipe-end-meta

name	haipipe-end-meta
description	MetaFn specialist -- designs/reviews the model-metadata-lookup function in an Endpoint_Set. One of 5 inference Fn-types. Called by /haipipe-end when intent references MetaFn, model metadata, model card, or `meta`.
argument-hint	[verb] [use_case] [args...]
allowed-tools	Bash, Read, Write, Edit, Grep, Glob
metadata	{"version":"0.1.1","last_updated":"2026-07-08","summary":"MetaFn specialist — design and review of the model-metadata-lookup function in an Endpoint_Set."}

Skill: haipipe-end-meta

Per-Fn-type specialist for MetaFn — the inference function that returns model metadata (model name, version, description, signature) without running prediction. One of the 5 inference Fn-types inside an Endpoint_Set. See ref/concepts.md for MetaFn semantics.

Verb axis: design | review | list | concepts Use case: each MetaFn impl is scoped to ONE product use case (CGMDecoder, SMS, Bandit, WeightLoss, …). design and review take a use_case argument.

---

Commands

/haipipe-end-meta                                       -> show MetaFn ref (concepts mode)
/haipipe-end-meta concepts                              -> same
/haipipe-end-meta list                                  -> list known use-case impls in fn_meta/
/haipipe-end-meta design <use_case> [endpoint_set]      -> scaffold a new MetaFn for <use_case>
/haipipe-end-meta review <use_case> [endpoint_set]      -> structural audit of one use-case impl

Use cases (concrete impls in code/haifn/fn_endpoint/fn_meta/)

AutoMetaFn (conceptual pattern, not a file)                                  generic auto-built MetaFn
BanditSMS_v250225                           SMS bandit
BanditSMSnNudge_v0620                       SMS+Nudge bandit
CGMDecoder_v260101                          CGM decoder (SageMaker)
CGMDecoder_Databricks_v260101           🚩  CGM decoder (Databricks variant)
MimicMortality_v260611                  🆕  MIMIC-IV mortality (binary classification)
SMSnNudge_v0620                             SMS+Nudge
SMSR2_13Messages                            R2: 13-message lineup
SMSR3_20Messages_SMSPersonalizeContent      R3: 20-message lineup with personalization
WeightLossMultiLabel_v260305                weight-loss multi-label
WeightLossMultiLabel_Af1M_v260310           weight-loss (Af1M variant)
WeightLossMultiLabel_OldFormat_v260318      weight-loss (legacy format)

🚩 = target-specific variant (Databricks)
🆕 = added in REACH-SPACE (2026-06)

If <use_case> is omitted, the skill should Bash("ls code/haifn/fn_endpoint/fn_meta/") and ask the user to pick.

---

Dispatch Table

Verb       Reads
---------- ------------------------------------------------------------------
design     ref/concepts.md
           ../haipipe-end/fn/fn-design.md
           ../haipipe-end/ref/0-overview.md   (Endpoint_Set + pipeline context)
review     ref/concepts.md
           ../haipipe-end-endpointset/fn/fn-review.md
concepts   ref/concepts.md  (only)

---

Step-by-Step Protocol

Step 0: Read ref/concepts.md — MetaFn semantics, expected I/O, registry pointers. Step 1: For design, also read ../haipipe-end/fn/fn-design.md + ../haipipe-end/ref/0-overview.md. For review, also read ../haipipe-end-endpointset/fn/fn-review.md. Step 2: Execute the procedure scoped to MetaFn (do NOT touch other Fn-types). Step 3: Emit the structured tail (umbrella parses this):

status:    ok | blocked | failed
summary:   2-3 sentences on what was designed/reviewed for MetaFn
artifacts: [paths to MetaFn files created or modified]
next:      suggested next command (typically: review, then sibling Fn-type)

---

Scope

Owns:

• MetaFn concept ref (ref/concepts.md)
• MetaFn design + review behavior scoped to ONE Fn-type
• MetaFn-specific files inside code/haifn/fn_endpoint/ and the Endpoint_Set

Does NOT own:

• Other 4 Fn-types — sibling skills -trig, -post, -src2input, -input2src
• Whole-artifact verbs (package / test / dashboard) — /haipipe-end-endpointset
• Deployment to any target — /haipipe-end-deploy-*

If a design fails because of an Endpoint_Set issue, escalate to /haipipe-end-endpointset review rather than patching here.

Builder examples (ref/examples/)

Three real builders are provided as reference — copy the closest one:

ref/examples/
├── a1_build_metafn_cgm_databricks.py        CGM (Databricks, time-series forecast)
├── a1_build_metafn_weight_multilabel.py     Weight (SageMaker, multi-label classification)
└── a1_build_metafn_mimic_mortality.py       MIMIC (binary classification, source_tables format)

All follow [BOILERPLATE]/[CUSTOMIZE] convention with region markers. See ref/concepts.md "Builder Examples" section for the comparison table.