khe-brain-benji
Answers design and product questions from Benji Taylor's published work, with the source named.
npx skills add HKTITAN/skills --skill khe-brain-benjiCredit
This is Harshit Khemani's reading of Benji Taylor's public writing and public skills. Benji Taylor did not write it, review it or endorse it. Read him directly at benji.org and use his own skills from github.com/benjitaylor/skills. When this skill and one of his pieces disagree, the piece is right. Every sentence here is a paraphrase; the few quotes are under 15 words each.
What it does
- Answers a question with the claim, the piece it comes from and its URL.
- Pressure-tests a design, a product decision or a piece of faux UI against his stated positions, one line per position.
- Says "he has not written about this" when the corpus is silent, and lists what was not read.
- Reports both sides where two pieces disagree or a view moved over time, with dates.
- Keeps each claim in its context: faux product UI, the Family wallet, the Honk messenger, a React library, or working with agents.
- Never builds. It names
stageshoworkhe-uiand stops.
How it runs
Used
/khe-brain-benji Would Benji ship this hero? The whole dashboard sits in a card and every row has real text.The gold reply for that hero, whose brief there adds that the status flips with no cursor, in full from examples/pressure-test-faux-ui.txt:
Against the positions in the Stageshow skill this hero breaks three and holds one. The largest is the frame: a whole window shrunk into a card with even margins is the failure he names boxed (Stageshow, SKILL.md).
Breaks: the full dashboard sits whole inside the card; a hero is cropped 10 to 40 percent off the stage and fills at least two thirds of its width (Stageshow, references/framing.md).
Breaks: every row carries real text; a scene usually holds 3 to 7 real strings, with placeholder lines for the rest (Stageshow, references/craft.md).
Breaks: the status flips with nothing causing it; a scripted change with no cursor reads as automatic (Stageshow).
Holds: one accent on one neutral family of surfaces (Stageshow, references/craft.md).
Source: Stageshow, SKILL.md, https://github.com/benjitaylor/skills
Source: Stageshow, references/framing.md, https://github.com/benjitaylor/skills
Source: Stageshow, references/craft.md, https://github.com/benjitaylor/skills
Source: Stageshow, https://benji.org/stageshow
Not in his writing: whether the real dashboard behind this picture is well designed; these are rules for a picture of a product on a page. To rebuild the hero use `stageshow`, which is his own skill.When to use it
Use it when you want an answer from a known source in place of an average of everything: before quoting him, when someone says "Benji says" and you want the piece, or when a design should be checked against a stated standard. Do not use it to make the thing. For faux product UI his own stageshow skill is the source and the tool; this skill only tells you what it says and where.
Invoke with
/khe-brain-benji [question | test <design or decision> | who said <line> | sources]
Or ask in your own words: "what would Benji say about", "would Benji ship this", "who said that".
Before and after
Before: "what would Benji say about confetti on the send button?" returns four confident paragraphs in his voice, a number he never gave and no link.
After: three sentences that place confetti on a rare moment and name the wallet it was said about, each ending (Family Values); then Source: Family Values, https://benji.org/family-values; then Not in his writing: and the part no piece answers.
Example prompts
- What would Benji say about showing every feature on the first screen?
- Would Benji ship this hero? The whole dashboard sits in a card and every row has real text.
- Who said "pointing beats describing", and where?
Verify
node scripts/brain-check.mjs reply.txt --gate # a reply, saved as text
node scripts/brain-check.mjs examples --gate # the gold replies
node scripts/brain-check.mjs --self-test # the gate against its own fixtures, and the corpus against its sourcesbrain-check.mjs names every fault as a line. --gate ends brain-check passed only when every never-intentional line is at 0, exits 1 with the counts otherwise, and exits 2 with inconclusive when it could not look.
What it proves: the reply opens with the answer, has at least one Source: line, cites only titles that are in references/sources.md with the right URL, lists every piece it cites, gives no sentence marked as his without a piece, keeps quotes under 15 words, uses no hedge word and no "he would", and ends with Not in his writing:. The self-test also proves that every citation in the theme files names a piece in the sources table.
What it does not prove: that a cited piece says what the sentence claims, that a quote is word for word, that a rule was kept in its context, or that the corpus is current. It matches titles, never meaning. Passing is necessary, not sufficient.
What is inside
- SKILL.md: the one idea, the workflow, the table from theme to file, the hard rules and the output format.
references/: the corpus by theme (restraint-and-detail.md,framing-and-composition.md,motion-with-a-cause.md,components-and-states.md,delight-and-play.md,verification-and-rehearsal.md,agents-and-tools.md,taste-and-practice.md), each claim one line with its title and URL;sources.md(28 pieces, dates, what was not read, how to refresh);disagreements.md;silent.md;verify.md,review.md,teaching.md; andgotchas.mdandpov.md, which are yours to edit.examples/: six gold replies: a direct question, two pressure-tests, a silent topic, a disagreement and a "who said that".teaching/: eleven bad/good pairs, each named for the gate line its bad half raises. They are the gate's tests.scripts/:brain-check.mjs(the gate). Node 20 or newer, no dependencies.- LICENSE: MIT, for this skill's own text and code. His writing and his skills stay his, under their own terms.