boost-skill-cli 1.2.22__tar.gz → 1.2.23__tar.gz

boost_skill_cli-1.2.23/.vale/styles/boost/AIWriting.yml

@@ -0,0 +1,69 @@
+extends: existence
+message: "'%s' is a common LLM writing tell — rewrite it plainly."
+level: warning
+ignorecase: true
+# Vocabulary that marks generated prose, drawn from Wikipedia's
+# "Signs of AI writing" (WP:AISIGNS) and confirmed against boost's own docs:
+# all 13 files the prose-lint workflow checks scored ZERO hits on this list
+# before it was added, so it gates new drafts rather than grandfathering old
+# ones. Scope is text only — Vale never looks inside code spans or fences, so
+# this file's own examples in `backticks` and any CLI output are unaffected.
+#
+# Two classes of tell are deliberately NOT listed here, because a linter reads
+# tokens and these are properties of a whole passage:
+#   * em-dash density — one is punctuation, nine on a screen is a signature.
+#   * rule-of-three parallelism, and closing lines that assert a thing's
+#     significance instead of stating what it does.
+# style/README.md ("Writing") carries those, since they need a human read.
+tokens:
+  # puffery and promotional register
+  - boasts?
+  - nestled
+  - groundbreaking
+  - renowned
+  - cutting[- ]edge
+  - state[- ]of[- ]the[- ]art
+  - game[- ]chang(er|ing)
+  - best[- ]in[- ]class
+  - seamless(ly)?
+  - vibrant
+  - rich tapestry
+  - tapestry
+  # inflated significance
+  - testament to
+  - pivotal
+  - stands as
+  - serves as
+  - plays a (vital|key|crucial|significant) role
+  - underscor(e|es|ing)
+  - highlight(s|ing) the importance
+  - indelible mark
+  - ever[- ]evolving
+  - evolving landscape
+  # filler analysis and hedged attribution
+  - valuable insights?
+  - it('s| is) worth noting
+  - experts (argue|say|agree)
+  - industry reports?
+  - observers have
+  # LLM-favoured connectives and verbs
+  - delve
+  - delving
+  - meticulous(ly)?
+  - intricac(y|ies)
+  - myriad
+  - plethora
+  - realm of
+  - navigate the
+  - foster(ing)?
+  - resonate(s|d)? with
+  - align(s|ed)? with
+  # negative parallelism
+  - not just .{1,40} but
+  - not only .{1,40} but also
+  # essay scaffolding
+  - in conclusion
+  - overall,
+  - moreover,
+  - furthermore,
+  - additionally,

{boost_skill_cli-1.2.22 → boost_skill_cli-1.2.23}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: boost-skill-cli
-Version: 1.2.22
+Version: 1.2.23
 Summary: boost — Homebrew for AI coding skills
 Author: Jonathan Reyes
 License: GNU GENERAL PUBLIC LICENSE

{boost_skill_cli-1.2.22 → boost_skill_cli-1.2.23}/boost_cli/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '1.2.22'
-__version_tuple__ = version_tuple = (1, 2, 22)
+__version__ = version = '1.2.23'
+__version_tuple__ = version_tuple = (1, 2, 23)
 
 __commit_id__ = commit_id = None

{boost_skill_cli-1.2.22 → boost_skill_cli-1.2.23}/boost_skill_cli.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: boost-skill-cli
-Version: 1.2.22
+Version: 1.2.23
 Summary: boost — Homebrew for AI coding skills
 Author: Jonathan Reyes
 License: GNU GENERAL PUBLIC LICENSE

{boost_skill_cli-1.2.22 → boost_skill_cli-1.2.23}/boost_skill_cli.egg-info/SOURCES.txt

@@ -24,6 +24,7 @@ noxfile.py
 pyproject.toml
 setup.cfg
 sonar-project.properties
+.vale/styles/boost/AIWriting.yml
 .vale/styles/boost/Terminology.yml
 .vale/styles/config/vocabularies/boost/accept.txt
 boost_cli/__init__.py

{boost_skill_cli-1.2.22 → boost_skill_cli-1.2.23}/style/README.md

@@ -76,6 +76,68 @@ each `.glass`, `.cap`, `.stat`, and `.window`. The CSS reads those in a radial
 gradient. No JS → the variables stay unset → the spotlight simply never shows,
 and the hover lift still works.
 
+## Type
+
+Both stacks are system fonts, so the system fetches nothing and there is no
+flash on load.
+
+A page may justify one display face. Three things then bite, and all three fail
+silently — the page renders, it just renders wrong.
+
+| Trap | What happens | The fix |
+|------|--------------|---------|
+| Enumerated weights | `wght@400;500;600` loads those three. `font-weight: 700` is then *synthesised*: the browser smears the 600 outline. | Request the axis: `wght@400..900`. |
+| Missing italic axis | `font-style: italic` shears the roman instead of loading the italic. On a high-contrast face a true italic is a different alphabet, so the fake reads immediately. | `ital,wght@0,400..900;1,400..900`, plus `font-synthesis: none` so the fallback is visible rather than silent. |
+| Display weight on a dark ground | A high-contrast face at 500 has hairline strokes. On `--bg` they dissolve, and under `background-clip: text` they take the gradient with them. | Set display type at 700+. Specimen sheets flatter 400-500 on white; this ground is not white. |
+
+To check a face actually loaded rather than trusting the render, compare advance
+widths in the console — a real italic sets narrower than a sheared roman:
+
+```js
+const w = (style, weight) => {
+  const c = document.createElement('canvas').getContext('2d');
+  c.font = `${style} ${weight} 100px "Your Face"`;
+  return c.measureText('remembered').width;
+};
+w('normal', 700);   // 612.4
+w('italic', 800);   // 576.7  <- a real italic cut, not a slant
+```
+
+### Gradient on type
+
+`.grad-text` paints only where a glyph is, so weight is a contrast control here
+rather than a style choice. Thin strokes give the ramp no area, and the text
+fades toward whichever stop sits closest to the ground. Anything wearing it
+wants 600+; `.eyebrow` (700) and `.stat b` (800) are the reference.
+
+## Writing
+
+Prose is part of the system. `.vale/styles/boost/AIWriting.yml` enforces the
+mechanical half of this and runs in the `prose-lint` workflow over every
+contributor-facing Markdown file, including this one. All 13 scored zero hits
+before the rule landed, so it gates new drafts rather than grandfathering old
+ones.
+
+Vale reads tokens, so it catches vocabulary and nothing else. These three are
+properties of a whole passage and need a human:
+
+**Em-dash density.** One is punctuation. Nine on a screen is a signature. The
+rule is not abstinence — this repo's own docs use them well — it is noticing
+when every other sentence has grown a subordinate clause instead of a full
+stop. A draft that arrives at ~25 per page is not written, it is generated.
+
+**Rule of three.** Three-item lists and three-clause sentences are the default
+cadence of generated prose. One is fine. Three sections in a row that each
+open with three parallel clauses is a tell.
+
+**Significance closers.** A sentence whose job is to assert that the preceding
+sentence mattered. `This is the part that really matters`, `that is the whole
+point`, `it is worth noting that`. Cut them and nothing is lost, which is the
+diagnostic: they carry no information the reader did not already have.
+
+Two habits that replace all three: say what a thing does rather than what it
+represents, and prefer the shorter of two accurate sentences.
+
 ## Accessibility
 
 - **`prefers-reduced-motion`** disables the aurora drift, the breathing badge,

{boost_skill_cli-1.2.22 → boost_skill_cli-1.2.23}/style/boost.css

@@ -68,7 +68,29 @@
   --term-bg:  #0d1024;
   --term-bar: #171a33;
 
-  /* --- type --- */
+  /* --- type ---
+     Both stacks are system fonts on purpose: the system fetches no webfont, so
+     there is no flash, no third-party request and no CSP entry to maintain.
+
+     ADDING A DISPLAY FACE (the one deviation a page may justify) has three
+     failure modes, all silent — the page renders, it just renders wrong:
+
+       1. Request the WEIGHT AXIS, not a few weights. `wght@400;500;600` loads
+          exactly those; `font-weight: 700` then gets SYNTHESISED — the browser
+          smears the 600 outline and calls it bold. Variable syntax is
+          `wght@400..900`, and one file covers the range.
+       2. Request the ITALIC AXIS separately, or `font-style: italic` is faked
+          by shearing the roman. On a high-contrast face (a didone, say) a real
+          italic is a different alphabet, not a slanted one, so the fake is
+          obvious. The axis is `ital,wght@0,400..900;1,400..900`, and
+          `font-synthesis: none` turns the silent fallback into a visible one.
+       3. High-contrast faces need WEIGHT to survive dark grounds. A didone's
+          hairlines at 500/90px on --bg vanish into the background, and under
+          background-clip:text they take the gradient with them (see .grad-text).
+          Set display type at 700+, not at the 400-500 a specimen sheet shows.
+
+     Measured while building a page on this system: 500 -> 700 was the whole
+     difference between a headline that dissolved and one that held. */
   --mono: ui-monospace, "SF Mono", SFMono-Regular, Menlo, Consolas, monospace;
   --sans: -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, "Helvetica Neue", Arial, sans-serif;
 
@@ -221,7 +243,13 @@ body::after {
    One instance per page, on the thing the page is actually about.
 
    background-clip:text zeroes the text fill, which also zeroes the selection
-   highlight, so ::selection has to put a colour back. */
+   highlight, so ::selection has to put a colour back.
+
+   It also means the gradient is only ever visible where a GLYPH is. Weight is
+   therefore a contrast control here, not a style choice: thin strokes give the
+   ramp almost no area to paint, and the text fades toward whichever stop is
+   darkest against the ground. Anything wearing .grad-text wants 600+ — the
+   .eyebrow (700) and .stat b (800) already do, and are the reference. */
 .grad-text {
   background: var(--grad);
   -webkit-background-clip: text; background-clip: text; color: transparent;