agentleFS
vivian-chivivian-chi/growing-brand/skills/brand-memory/SKILL.md

brand-memory

Skill1 starsChanged 4 months agoMITView on GitHub

Use during any design or visual work — building or styling web pages, components, slides, posters, social images, emails, documents, data visualizations — and whenever the user reacts to something visual (picks one option over another, rejects a direction, asks for a visual change like "warmer" or "less rounded", approves a result, or states a design rule). Loads the project's brand memory from brand/ before design choices are made, quietly records the user's design judgments as they naturally occur, and automatically distills repeated judgments into the brand files. The user never has to invoke anything — this layer rides along inside normal design work.

2 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions2 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 2

    1. Lucas-Grilli/SMM_OS★ 2same name
    2. akachandan1/proofstack★ 0same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repository2 other files it gives its agents.

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 5 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    The loop: APPLY → CAPTURE → DISTILL: All three happen inside normal design work.

    line 13
  2. Very specific:

    Read brand/core.md, brand/tokens.md, brand/moves.md, and the undistilled tail of brand/journal.md (entries after the Last distilled marker …

    line 24
  3. Distinctive:

    Write silently — don't interrupt the flow to announce a journal entry.

    line 50
  4. Distinctive:

    Trigger — no permission, no ceremony: after appending, if undistilled entries ≥ 10, distill in the same session.

    line 57
  5. Distinctive:

    Read all undistilled entries plus the three brand files.

    line 59
  6. Distinctive:

    Demotion works too: if new entries contradict a confirmed rule ≥ 2 times, move it to Tensions and say so.

    line 69
  7. Very specific:

    Update the journal marker: Last distilled: <date> · through entry <N>.

    line 70
  8. Very specific:

    Never write into core.md without an evidence trail — provenance always.

    line 80

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

brand-memory — the layer that grows a brand while you design

Nobody writes their taste down in advance. It leaks out one small judgment at a time — "not this one", "warmer", "less rounded", "yes, that's it." Normally those judgments evaporate when the session ends. This skill catches them while the user works and grows them into a brand system the user never had to author.

The loop: APPLY → CAPTURE → DISTILL. All three happen inside normal design work. No practice routine, no review sessions, no rating ceremonies — the user just works.

0 · When this activates

  • Any design/visual making, in any conversation — even one that didn't start as design work.
  • Any user reaction to something visual you produced.
  • If the project has a brand/ folder (with journal.md in it): apply + capture, silently. (If brand/ belonged to something else at init time, the folder is growing-brand/ instead — check for growing-brand/journal.md too and use that folder throughout.)
  • If it doesn't: the first time a real design judgment happens, offer once, in one line: "Want me to remember your design taste as we go? I'll keep it in brand/ — plain files, yours. (yes / no / /brand-init to seed it from your existing work)". If yes, create brand/ from the brand-init skill's templates and continue. If no or no answer, drop it — don't ask again this session, work normally.

1 · APPLY — before making design choices

  1. Read brand/core.md, brand/tokens.md, brand/moves.md, and the undistilled tail of brand/journal.md (entries after the Last distilled marker — the freshest signal, not yet promoted).
  2. Precedence when they speak to the same choice: core > tokens > moves > journal tail.
  3. Wear the brand by default. Don't ask permission to follow the user's own recorded taste, and don't recite the rules back — just make choices that obey them.
  4. On conflict — the user asks for something a confirmed rule forbids — say it in one line: "Your brand notes say [rule] — override here?" Either answer is a judgment: an override is an exception worth journaling, or the first evidence the rule is wrong.
  5. Lines marked (observed) or (seeded) are weaker than lines carrying journal evidence — bend those without asking.

2 · CAPTURE — record judgments as they happen

What counts (each becomes one journal line):

TypeLooks like
[picked]chose one option over another — "the second one"
[rejected]turned a direction down — "no purple", "not this"
[corrected]asked for a visual change — "warmer", "smaller", "less rounded"
[approved]accepted a result — "yes", "ship it", "love this"
[stated]declared a rule outright — "we never use stock icons"

How to record: append to brand/journal.md under ## Entries, numbered sequentially:

7. 2026-07-05 [corrected] buttons rounded-full → 4px — "we're not a toy" (job: pricing page)
  • Quote the user's actual words whenever they gave a reason. Their words are the data — don't paraphrase them into blandness.
  • One line per judgment; always name the job (what was being made).
  • Write silently — don't interrupt the flow to announce a journal entry.
  • A user hand-editing any brand/ file is the strongest signal there is — treat the edit as [stated].

Signal discipline — do NOT record: functional fixes (broken link, typo, bug), pure content/copy edits, indecision or thinking-out-loud, my own choices the user never reacted to, anything personal or sensitive. If one session produced a pile of judgments, keep the strongest ~8 — a journal padded with noise distills into a mushy brand.

3 · DISTILL — automatic summarize-upward

Trigger — no permission, no ceremony: after appending, if undistilled entries ≥ 10, distill in the same session. Also distill when loading a brand whose backlog is already ≥ 10.

  1. Read all undistilled entries plus the three brand files.
  2. Find repetition: the same preference or aversion appearing ≥ 2 times across different jobs.
  3. Promote each pattern to the level it belongs to:

    • a measurable (color, type, spacing, radius, density) → tokens.md
    • a recurring element, composition habit, or "never appears" → moves.md
    • an identity-level always/never, ≥ 3 consistent occurrences across ≥ 2 jobs → core.md — and keep core small (~15 rules max; when in doubt, it's a token or a move, not core)

    Two shortcuts past the repetition bar: a single entry that confirms an existing (observed)/(seeded) line upgrades it — drop the marker, cite the entry. A [stated] rule goes straight to its level — the user said it outright; cite the entry.

  4. Provenance on every promoted or updated line: — evidence: entries 5, 11. A rule that can't cite its entries doesn't get written.
  5. Contradictions are never resolved by fiat. First try a context split ("4px radius on product UI, 12px on marketing pages" — different jobs, both true). If no clean split, put both sides in core.md's Tensions section with their entries and leave it open — later evidence settles it.
  6. Demotion works too: if new entries contradict a confirmed rule ≥ 2 times, move it to Tensions and say so.
  7. Update the journal marker: Last distilled: <date> · through entry <N>. Entries stay in the journal — they are the permanent record.
  8. Tell the user what changed in ≤ 3 short lines — e.g. "Noticed: you've rejected pure black twice on different jobs — added 'ink #1A1F1A, never #000' to your tokens." If they push back, that pushback is a judgment: record it and revert the promotion.

4 · STATUS — when asked

"What do you know about my brand?" / /brand-memory → a short summary: what's in core / tokens / moves (counts + the newest rules), patterns currently forming (seen 2×, not yet promoted), open tensions. No advice, no ceremony.

Boundaries

  • brand/ belongs to the user: plain markdown in their repo, theirs to edit or delete. Nothing is stored anywhere else; nothing leaves the project.
  • Never write into core.md without an evidence trail — provenance always.
  • One project = one brand. (Different brand → different project/folder.)
  • This layer doesn't generate designs by itself — it makes whatever gets generated wear the user's accumulated taste, and gets sharper every time they react.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.