[firejune]

Closes #12. Owner: "딴 머리 치마에 물리도 넣어줘" → narrowed to the two front side locks + the skirt ("땋은머리는 사실 안보여서 앞머리 양옆 두갈래만").

What

• New step 9c "Let the hair and the skirt answer": a cloth block (bottomwear: 3 chains × 3 links, swing scales [1.2, 1, 1.2]; front hair: 1 chain × 3 links, edge strength 100), rig-parts propose --cloth config.json --write-config, then rig-parts build … --loop --lead-in 1. Producer-first: the recipe drafts everything (27 physics + 3 transform constraints: the side locks as lockt0_/lock1t0_, the fringe bang_*, skirt0_..2_ + skirt_swing under hip); their idle keys are removed; bones 70 → 46. The bun is back hair above the shoulders, which the recipe refuses — it keeps its key (the owner said the braid is not visible).
• scripts/hello.ts --no-skirt-keys: in 9c the skirt is left to physics. Step 9b is byte-identical to PR feat: step 9b — a hello the producer builds: hm? → heh → elbow-down wave → weight-shift settle (#9) #11 (without physics the keys are still needed: thigh_l|bottomwear reads 48 px without them).
• Two corrections the draft needs on this example, shown as commands in the step: the bun track's amplitude 0.8 → 1 (CHECK_CHAIN_LAG reads neck > head_ctl > bun_ctl as one chain once the siblings' keys are gone — the recipe as drafted fails check here; filed upstream) and lock edge strength 100 (the default 50 ghosts 122 px at the "hm?" head turn).
• check-adapters.ts runs 9c and asserts 27 physics + 3 transforms and the named constraints. REFERENCE: constraint table, build lines, bars, angle/tip readings, wind reading, timings, versions.

Build lines: loop physics: 27 physics constraint(s) left out …; CLOTH_SETTLES: WARN — 9 physics chain(s) …: 4 of them settle before the loop's length (skirt trees 1.8–2.0 s, hair does not settle) — reported as is; ANIMATIONS_DO_NOT_FOLD: PASS on and off; check: PASS; 11 of 12, 1 skipped, 5 warnings (SEAM_OVER_MOTION, CLOTH_SETTLES); build: PASS. Without --lead-in, idle.gif jumps at the wrap (max |d| 209); with --lead-in 1, 0.

Bars (stated first; rigc render 24 fps, physics on from rest): all 15 seam pairs within gap ≤ 25 / ghost ≤ 50 on both idle and hello (closest face|hair_front 7/47 on the hello); pass-through 0/81 idle, 0/97 hello; sleeve_r0 4.00°.

What it moves: hello — locks trail the head ≈ 5 px, fringe ≈ 10 px at the "hm?" turn, skirt swing ≈ 3° / hem ≈ 7 px at the settle. Idle — after the 2 s settle the hair is near still (0.2–0.4 px) and the skirt < 0.01 px: by the parts guide's design a breath does not drive physics; the world does (wind reading: locks 9–10 px, hems 8–13 px under wind (2, 0)). Candidly, the idle now looks quieter than step 8's keyed 3.6° lock sway; a page-side wind through play's environment() would make it visible — not in this PR.

Reproduction: README as written from a clean folder, bun add resolved parts 1.49.1 (published mid-task): every expected line held; hello.anim.json, config.json, skeleton.json, skeleton.model.json, skeleton.png, idle.gif byte-identical to the 1.48.0 build measured. check-adapters.ts exit 0 (99 links, 0 failures).

Not measured: the hello layered over the idle on two tracks with physics (rigc render reads one animation); the player's own physics vs rigc's render.

🤖 Generated with Claude Code