docs(skills): fix zoom-through total and drop redundant keyword bracket in cut-the-curve

[imprimisxo]

What

Two documentation fixes to the cut-the-curve skill, applied to both mirrored copies (.agents/skills/ and .claude/skills/, byte-identical at HEAD):

1. §1 Zoom-Through: Total ≈ 0.4s → Total ≈ 0.7s
2. Frontmatter description: dropped the trailing 25-tag keyword bracket (949 → 663 chars)

Plus one question in "Open question" below that I did not change, because I can't tell which number is intended.

No behavioural or render changes.

Why

1. §1's stated total contradicts its own table

§1 says Total ≈ 0.4s, but the table directly beneath gives Exit 0.2s and Entry 0.5s. Those phases are sequential — the hard swap sits between them — so the total is 0.7s.

§2 Inverse Zoom-Through carries the same durations (~0.2s exit, ~0.5s entry) and states it correctly: Total ≈ 0.7s (30% exit / 70% entry). 0.2/0.7 ≈ 29%, 0.5/0.7 ≈ 71% — §2's arithmetic checks out, §1's does not.

If 0.4s was the intended budget then the table's durations are what need changing instead; happy to flip the fix.

2. The keyword bracket costs 285 chars and adds ~nothing

[depth, zoom, inverse-zoom, scale-sign, mirrored-zoom, rack-focus, pacing, velocity,
cut-the-curve, waterfall, stagger, cascade, kinetic-text, title-card, segment-opener,
nudge, slide, easing, group-motion, z-depth, motion-graphics, cinematic, transition,
blur, directional-continuity]

285 of the description's 949 characters. Checking every tag against the skill text:

• 17 of 25 already appear verbatim in the description prose above it — zoom, inverse-zoom, scale-sign, rack-focus, pacing, velocity, cut-the-curve, waterfall, stagger, cascade, kinetic-text, title-card, segment-opener, nudge, slide, transition, blur
• 1 — easing — is present as "mirrored power4 eases"
• 7 — depth, mirrored-zoom, group-motion, z-depth, motion-graphics, cinematic, directional-continuity — appear nowhere in the skill, neither in the description nor anywhere in the body

So the bracket restates terms already present, plus seven the skill never uses.

Why the 285 chars are worth reclaiming

Claude Code's skill listing is hard-capped at ~30,000 characters. Measured across 127 recorded listings in my own session transcripts: at 52 skills the listing is 7,986 chars, but every listing from 67 skills upward pins between 29,934 and 30,050 — 67, 85, 120, 142, 171 and 186 skills all land at ~30k. Past the cap Claude Code keeps each skill's name and drops descriptions from the tail wholesale.

That makes description length zero-sum across a user's entire skill set, not just a per-skill cost. In my install cut-the-curve's 949-char description is the 3rd largest of 85 skills, and 31 skills are already reduced to bare names with no description at all — including seam-craft and oversized-cursor, two of cut-the-curve's own siblings in the motion-doctrine router graph. A sibling stripped to a bare name can't be matched on intent, which is the failure mode the tags were presumably added to prevent.

How

Two single-line edits per file, applied identically to both copies so they stay in sync. Nothing else touched — no table values, no mechanics, no code templates.

Open question — "25–30% of travel" doesn't reconcile with the stated fade durations

Left unchanged; I don't know which number is authoritative.

§3 (The fade trick):

exit opacity completes at ~25–30% of its travel (fade ≈ 0.18–0.3s vs motion 0.3–0.34s)

§4 (Waterfall Cut table): exit 0.34s power4.in, exit fade 0.18s, "word gone by ~25–30% of travel".

Reading "travel" as distance: under power4.in the distance fraction is (t/T)^4. With t = 0.18s, T = 0.34s → (0.529)^4 ≈ 0.08. The word is 8% along, not 25–30%. To be gone by 25% of distance the fade would need t = T · 0.25^0.25 ≈ 0.24s (0.25s for 30%).

Reading "travel" as elapsed time: 0.18 / 0.34 ≈ 53% — also not 25–30%.

§3's own ranges straddle the same gap: fade 0.18–0.3s against motion 0.3–0.34s is 53–100% of the time, i.e. ~8–100% of the distance.

So either the fade durations want lengthening toward ~0.24s, or the "25–30%" figure wants restating, or "travel" means something specific worth spelling out. Glad to send that fix once I know which.

Test plan

• Unit tests added/updated — n/a, documentation only
• Manual testing performed — verified both files on the branch are byte-identical to the intended content; diff against main is 2 files, +4 −4, confined to the two lines described
• Documentation updated (if applicable) — this PR is the documentation change

Found while auditing skill descriptions for listing-budget cost.