@wildmason/aegis 1.15.0 → 1.18.0

package/CHANGELOG.md

@@ -10,13 +10,124 @@ page; both were written at the time, so the entries are sourced rather than
 reconstructed from memory. Anything before 1.8.0 lives in `git log` only, and is
 deliberately not summarised here rather than guessed at.
 
-**Releases from 1.12.0 onward carry no git tag.** They were published through the
-GitHub OIDC workflow, which does not tag, so `git tag` stops at `v1.11.0` and the
-tag names before that are inconsistent (`1.4.4`, `v1.5.1`). Version links below
-are omitted where no tag exists rather than pointed at a URL that 404s.
+**Releases 1.12.0 through 1.14.1 carry no git tag.** They were published through
+the GitHub OIDC workflow, which does not tag, so `git tag` skips from `v1.11.0` to
+`v1.15.0`, and the tag names before that are inconsistent (`1.4.4`, `v1.5.1`).
+Version links below are omitted where no tag exists rather than pointed at a URL
+that 404s.
 
 ## [Unreleased]
 
+_No unreleased changes._
+
+## [1.18.0] - 2026-09-02
+
+### Added
+
+- `WmRichTooltip` — a peer to the text-only `WmTooltip` for generic Angular
+  markup. Consumers project a trigger marked `wmRichTooltipTrigger` and an
+  `ng-template` marked `wmRichTooltipContent`; the template renders in the
+  same anchored overlay shape with focus, hover, Escape, ARIA, scroll, and
+  clipping behavior. It accepts no raw HTML string, leaving any explicit
+  `[innerHTML]` sanitization decision with the consumer. Helm's Stacks
+  portfolio uses it to pair full branch refs and state swatches with the
+  topology legend, including omitted overflow branches.
+
+## [1.17.0] - 2026-09-01
+
+### Added
+
+- `.wm-kbd--on-fill` — the same key cap `.wm-kbd--on-accent` shipped as one
+  release ago, driven by `currentColor` instead of `--wm-color-accent-text` so
+  it is correct on any control that carries its own fill rather than on the
+  accent alone. On `.wm-btn--primary` `currentColor` **is**
+  `--wm-color-accent-text`, so the accent case computes to exactly what 1.16.0
+  produced and `check:accent-contrast` measures the same numbers.
+
+  The generality was not speculative. Helm's conflict-resolution row puts a cap
+  inside a Reject button filled with `--wm-color-danger` at 22% over a tinted
+  header. Measured on all twelve themes, a plain `.wm-kbd` there draws an edge
+  that separates from the button at **1.65:1 to 2.89:1 — below WCAG 1.4.11's
+  3:1 on every one of them**. The same cap under this modifier measures 3.55:1
+  to 4.43:1. Nothing about that button is unusual; the accent-only modifier
+  simply had no answer for it.
+
+### Deprecated
+
+- `.wm-kbd--on-accent`, in favour of `.wm-kbd--on-fill`. It still ships, as the
+  same declarations, because removing a published class is a breaking change —
+  and it had no consumers to break, which is precisely why the name was worth
+  correcting one release later rather than living with. Slated for removal at
+  2.0.0.
+
+## [1.16.0] - 2026-09-01
+
+### Added
+
+- `.wm-kbd--on-accent` — the key cap for an accent fill. The base `.wm-kbd` is
+  `--wm-bg-sunken` inside `--wm-border-input`: a recess in a *surface*, and both
+  tokens are wrong the moment the cap sits on `.wm-btn--primary`. The modifier
+  carries no fill of its own, so the cap's label **is** the button's label and
+  inherits the contrast that pair already guarantees; only its edge is new, drawn
+  as `--wm-color-accent-text` at 75% over whatever is behind. Mixing toward
+  `transparent` rather than toward `--wm-color-accent` is what lets one
+  declaration serve the rest fill and the hover fill both.
+- `check:accent-contrast` — measures `.wm-btn--primary`'s label on both of its
+  fills, the new cap's edge on both, and how far the hover fill sits from the rest
+  fill, across every theme. Its destructive twin has had a guard since
+  white-on-danger was caught failing 8 of 12 themes; the accent button is the same
+  shape and had none.
+
+  The separation bar is there because the first pass shipped without one and eight
+  hovers came out invisible: the derivation took the smallest lightness step that
+  cleared contrast, which is a hover nobody can see. It is expressed as a Weber
+  fraction, `|L_hover - L_rest| / L_rest >= 0.14`, because an ABSOLUTE distance
+  cannot separate the accepted values from the rejected ones at any threshold —
+  sorted by OKLab distance an accepted theme sits below a rejected one. Normalised
+  by the rest fill's own lightness they separate cleanly, and 0.14 is the low end
+  of the band occupied by the three hovers this repo already shipped and nobody
+  ever complained about.
+
+### Fixed
+
+- **`.wm-btn--primary:hover` failed WCAG AA on warm-light: white on
+  `--wm-color-accent-hover` `#2aa198` measured 3.16:1.** The cause was
+  structural rather than one bad value. On five themes the hover was the
+  palette's *second accent* rather than a lighter or darker step of the first
+  — Solarized cyan behind Solarized blue, Dracula pink behind Dracula purple,
+  Monokai orange behind Monokai magenta. A hue jump promises nothing about the
+  label riding on it, so the label's contrast was whatever the two palette
+  entries happened to be, and on hover it also missed APCA Lc 60 on nine themes,
+  as low as Lc 28.7. Nine hover values are now the accent's own hue and chroma at
+  a different OKLCh lightness, moved *away* from the label, far enough to clear the
+  separation bar as well as the contrast bars. Where sRGB cannot hold the accent's
+  chroma at the higher lightness — vampires-kiss and twilight-peach — the colour
+  is gamut mapped with hue and lightness held and chroma reduced, rather than
+  clamped per channel, which shifts hue.
+- `--wm-color-accent-text` on arctic-night, vampires-kiss and twilight-peach was
+  short of APCA Lc 60 on the rest fill by 0.1, 0.3 and 1.4. Each moved a little
+  further from its fill in OKLCh lightness — hue and chroma held, two of the
+  three by under five values out of 255.
+- `DESIGN_LANGUAGE.md` §13 claimed `.wm-kbd` was "used identically in Helm
+  (command palette) and Slate (sidebar search button)". Neither consumer used it
+  at all. Helm draws seven key badges from five component-local rules; Slate draws
+  none. `check:docs` gates class *existence* in both directions and cannot see a
+  false claim about who uses one, so this was found by looking.
+
+### Known limitation
+
+- **Four accents cannot reach APCA Lc 60 with any label, and are pinned rather
+  than repaired. All four clear WCAG on both fills** — WCAG is the bar this
+  package ships to and no theme is exempt from it; APCA Lc 60 is the bar it
+  reaches for, enforced the same way because a bar nothing enforces decays. A
+  pinned theme is behind an aspiration, not broken. Pure white and pure black bracket every achievable Lc, and on
+  posh-sandalwood (`#f92672`, best 40.7 dark / 68.8 white — but white measures
+  3.79:1 and fails WCAG outright), source-control-dark (`#3fb950`, 54.6),
+  crepuscular-sky (`#c09d06`, 55.7) and wildmason (`#82a682`, 57.8), neither end
+  reaches it. Clearing the bar means changing the accent itself, which is each
+  theme's brand colour. The shortfalls are recorded to one decimal in
+  `check-accent-contrast.mjs`, which fails if one drifts in either direction.
+
 ## [1.15.0] - 2026-09-01
 
 ### Added
@@ -231,6 +342,10 @@ are omitted where no tag exists rather than pointed at a URL that 404s.
 Releases before 1.8.0 are in `git log`. See also `wiki/products/Aegis - Health.md`
 for the decisions behind these changes, including the ones decided against.
 
-[Unreleased]: https://github.com/wildmason/aegis/compare/master...HEAD
+[Unreleased]: https://github.com/wildmason/aegis/compare/v1.18.0...HEAD
+[1.18.0]: https://github.com/wildmason/aegis/releases/tag/v1.18.0
+[1.17.0]: https://github.com/wildmason/aegis/releases/tag/v1.17.0
+[1.16.0]: https://github.com/wildmason/aegis/releases/tag/v1.16.0
+[1.15.0]: https://github.com/wildmason/aegis/releases/tag/v1.15.0
 [1.11.0]: https://github.com/wildmason/aegis/releases/tag/v1.11.0
 [1.9.0]: https://github.com/wildmason/aegis/releases/tag/v1.9.0