agentleFS
vivian-chivivian-chi/growing-brand/CLAUDE.md

growing-brand

CLAUDE.md1 starsChanged 4 months agoMITView on GitHub

An ambient brand-memory layer for AI-assisted design work. Audience = designers and design-engineers using AI coding tools. Assume users have never seen this repo's internals and never will — the tool has to work inside their normal flow. There is no test suite — the spec is the program and the AI is the runtime. So test by dry run: create a sandbox project with some CSS, execute /brand-init faithfully (real greps, real files), simulate ~12 judgments across ≥3 fake…

The file

We split this CLAUDE.md into the rules it gives an agent

1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Provenance always: No rule enters core/tokens/moves without citing journal entries.

    line 18

≈ “Common” is an estimate: at least 1 in 5 of the 28,029 CLAUDE.md 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
CLAUDE.md

CLAUDE.md — growing-brand (contributor side)

What this is

An ambient brand-memory layer for AI-assisted design work. Audience = designers and design-engineers using AI coding tools. Assume users have never seen this repo's internals and never will — the tool has to work inside their normal flow.

Audience rules (the reason this tool is shaped the way it is)

  • Users will not do daily practice, review sessions, or any routine. If a feature needs discipline, it's wrong.
  • Users will not run judging ceremonies (generate K options, rate them, write reasons). Their judgments already happen inside normal work — "not this one", "warmer", "approved" — the tool's only job is to catch them there.
  • Capture is ambient and silent; distillation is automatic (threshold-triggered, no permission asked) but visible in ≤3 lines and reversible — a pushback on a promotion is itself a judgment.
  • Plain language everywhere. Public files (README, SKILL.md, templates) use only: core / tokens / moves / journal.

Design principles

  • Zero runtime dependencies. The whole tool is markdown: two skills + four templates. No scripts, no node_modules, no server. The AI harness is the runtime. Keep it that way — any feature that needs code should justify itself hard.
  • Files are the product. The grown brand/ folder is the value; the skills are just the gardener. Files must stay readable, diffable, portable, and edit-friendly (hand edits = strongest signal).
  • Provenance always. No rule enters core/tokens/moves without citing journal entries. A rule that can't show its evidence doesn't get written.
  • Contradictions stay open. Context-split first; otherwise Tensions. Never resolve by fiat; demotion exists.
  • Core stays small. ~15 rules max. When in doubt, it's a token or a move.
  • The user's words beat paraphrase. Quote them in the journal.

Layout

.claude-plugin/plugin.json      manifest (validated with `claude plugin validate .`)
.claude-plugin/marketplace.json makes the repo directly installable as a plugin
skills/brand-memory/SKILL.md    the ambient layer: apply → capture → distill (the heart)
skills/brand-init/SKILL.md      one-time onboarding: scan + ≤3 questions + seed
skills/brand-init/templates/    core.md · tokens.md · moves.md · journal.md
example/                        fictional brand (Ridgeline) — generated by executing the
                                loop end-to-end once; regenerate it the same way if the
                                spec changes, don't hand-tweak it out of sync

How to test a change

There is no test suite — the spec is the program and the AI is the runtime. So test by dry run: create a sandbox project with some CSS, execute /brand-init faithfully (real greps, real files), simulate ~12 judgments across ≥3 fake jobs, let the distill trigger fire, and check: entry numbering, marker update, provenance on every promoted line, observed/seeded upgrades, [stated] fast-path, tension handling. Any ambiguity you hit while executing is a bug in the SKILL text — fix the text, not the run.

v1 candidates

Hook-triggered distill (SessionEnd), deeper init extraction, an optional render/validate floor for HTML deliverables, team/shared brands, journal compaction past ~200 entries.

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.