haipipe-nn-algo

name	haipipe-nn-algo
description	Layer 1 (Algorithm) specialist of haipipe-nn: defines the algorithm contract (model class, forward pass, loss, metric). Covers mlpredictor, tsforecast, tefm, tediffusion, bandit. 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 1 (Algorithm) specialist of haipipe-nn."}

Skill: haipipe-nn-algo

Layer 1 specialist. Owns the algorithm definitions in code/hainn/algo/. Algorithms are the model classes — forward pass, loss, metric — without any hyperparameter sweep or training loop logic. Called by the /haipipe-nn orchestrator; can also be invoked directly.

Function axis: dashboard | review | generate | test

---

Commands

/haipipe-nn-algo                       -> dashboard: registered algorithms
/haipipe-nn-algo dashboard             -> same
/haipipe-nn-algo review [class_name]   -> structural review of an algorithm class
/haipipe-nn-algo generate              -> scaffold a new algorithm class
/haipipe-nn-algo test [class_name]     -> run unit / smoke tests for the algorithm

---

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-tuner/
                 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)

generate reads the L2 (tuner) ref because a new algorithm must satisfy the tuner's expected __init__ signature for hyperparameter binding.

---

Step-by-Step Protocol

Step 0: Read ../haipipe-nn/ref/overview.md for the 4-layer map. Mandatory. Step 1: Parse args. Function vocabulary: dashboard | review | generate | test. Step 2: Read this skill's ref/concepts.md for L1 specifics. Step 3: Read the umbrella fn doc. Step 4: For generate, also read ../haipipe-nn-tuner/ref/concepts.md. Step 5: Execute, scoped to L1 (algorithm class only — no tuner / instance work). Step 6: Emit the structured tail (status / summary / artifacts / next).

---

Layer Scope

Owns:

• Algorithm class definitions in code/hainn/algo/
• Per-family conventions: mlpredictor, tsforecast, tefm, tediffusion, bandit
• Forward pass, loss, metric

Does NOT own:

• Hyperparameter sweep logic (/haipipe-nn-tuner)
• Trained instance weights (/haipipe-nn-instance)
• Multi-instance pipelines (/haipipe-nn-modelset)

Hand-off contract (L1 -> L2): Algorithm __init__ parameters become the tuner's search-space schema. Verify against ../haipipe-nn-tuner/ref/concepts.md before locking parameter names.