neu — build plan

The neural CLI (neu for adoption; the native tongue of the cyb terminal inside the ecosystem). A homoiconic client of the cell: parse → expand → seal, read via inf. Rides nox → bbg → cybergraph → inf — no new core deps. Spec: neural/specs/cli.md; substrate seam: neural/specs/binding.md; relation model: neural/specs/relation.md.

Each slice is one green test — the same discipline as the cyb cell.

✅ Slice 1 — speak a typed relation (DONE)

neu "cat is-a animal" → sigil-parse → dialect expands to [cat→animal, H(cat,animal)→is_a] → lands on a durable Cell → neu focus reads it back, the relation-word ranked by focus.

• workspace neural/rs/ (sigil, dialect, cli), path-deps to the sibling crates.
• 6 tests green, full stack compiles ~15s. Binary runs: two sentences land, is_a shows energy 2 (relations are ranked particles, not super-nodes), object merges across sentences.
• verified: RUSTC_BOOTSTRAP=1 cargo test --workspace.

✅ Slice 2 — the sentence is atomic (DONE)

Base + axon-type links now fold into one signal. Added Cell::cast(neuron, links) in cyb-core (the durable choke point); Cell::link is now a one-link cast. neu::speak casts the whole sentence at once.

• verified: new test sentence_is_one_atomic_signal — two links, but cell.len() grows by exactly 1 (one signal), and both the base and the axon-type link are present. neu cli 3/3 green.
• no regression: all 5 cyb-core Cell tests still pass after the link→cast refactor (chain, gossip, durable).
• resolves the open "atomic cast seam" — chose the Cell method so the durable log stays the one choke point.

Slice 3 — read is neural, not raw inf

Wrap the ~5 canned inf templates behind verbs: focus (particles by φ*), neighborhood W..X, path A..B, and the typed-relation join (neu ls --rel is-a → all X is-a Y via the axons join). Design once, reuse. Test: X is-a Y round-trips through the query, not through cell.axons().

Slice 4 — dialect scope + genesis vocabulary

--dialect D scopes resolution. Ship the bootloader relation-words (TRUE/FALSE + is-a/causes/part-of/contradicts + the image schemas). neu dialect ls/diff. Test: the same word resolves to different relations under two dialects; diff surfaces a collision.

Slice 5 — identity & proof-as-signature

Real neuron key; attach the H(secret)=address proof (mudra) to the signal instead of the fixed @you. Test: a signal carries a verifiable author; a forged author is rejected.

Slice 6 — the REPL

neu (no args) → the live loop: speak, watch φ* / lexicon move, read. Fuses the three faces. Test: a session of sentences converges the same state as the equivalent one-shots.

Later — the hero (waits on tru)

neu ask [--gaps] (motif-transition foresight) and neu prove (zheng) need tru's tri-kernel + the motif surface. Design is in cli.md; the engine is the MIND layer, post-testnet. Ship the write+read core first.

Open decisions (gate specific slices)

• graphlet vs motif — gates motif's output schema and ask (Slice ≥ hero). Settle the term (graphlet = form, motif = significant form) before motif.md is frozen.
• sigils.md reconciliation — the older markup-framed spec still lists obsolete future-work (negation → valence, weight → stake, typed-edges → relation.md); prune before the full sigil grammar lands (Slice 3+).
• atomic cast seam — Slice 2: add to Cell (touches cyb-core) vs build the signal in neu and apply via cell.graph. Prefer the Cell method — keeps the durable log as the one choke point.