agentleFS
kwakseongjaekwakseongjae/oh-my-design/.claude/skills/omd-add-reference/SKILL.md

omd:add-reference

Skill588 starsChanged 35 days agoMITView on GitHub

Reference catalog authoring compatibility gate. Core v2 cutover가 끝나기 전에는 신규 legacy DESIGN.md 쓰기를 차단하고 안전한 재개 조건을 설명한다.

The file

We split this skill 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:

    Never restore a legacy YAML marker, 13/15/16-section output, placeholder sentinels, or a second authoritative Markdown dialect as a …

    line 51

≈ “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

Expiry. This gate is a waiting state, not a policy. It lifts when the catalog adoption path lands — the T2-2 writer gate. Status, updated 2026-09-16 (the earlier "today: 없다" note was written 2026-09-02 and went stale):

  • graph-backed Core v2 writer exists — met. --adoption-target reference-catalog shipped with a four-canary isolated adopt → rollback → readopt rehearsal (2026-09-08, .omd/execution/2026-09-08/sprint90/core/).
  • every catalog reader accepts its package — not met. The web readers still parse legacy frontmatter and numbered sections (extract-tokens.ts, generate-css.ts, font-registry.ts, logos.ts).

Scope, narrowed by owner decision 2026-09-16. The second condition governs Core v2 packages, so it cannot govern a write that produces none. Two different operations were being blocked by one rule:

OperationProduces a Core package?Status
Create a new legacy reference, or a new Core catalog packageyes / adds migration debtstill refused
Refresh evidence on an existing reference — verification_v2, .verification.md, tokens.components state values, ## Proofno; the file stays legacy and stays in the same migration queueallowed

An evidence refresh does not create another legacy file — it restores what an earlier pass discarded from a file already queued for migration. It therefore falls outside this gate's stated purpose.

Still refused, with no one-off exception and no plausible fallback: creating a new reference, adding a brand, restoring a legacy YAML marker, emitting 13/15/16-section output, placeholder sentinels, a second authoritative Markdown dialect, or mutating generated counts and mirrors by hand.

omd:add-reference — Core v2 catalog cutover gate

Status: CORE_V2_CATALOG_WRITE_BLOCKED.

The historical reference catalog remains a dual-read migration input. It is not permission to create another legacy file. Do not write or update web/references/<id>/DESIGN.md, frontmatter tokens, generated counts, or mirrors.

On invocation, return the blocker and the proposed reference id/source list without writing. Authoring may resume only after a graph-backed reference metadata/evidence schema, deterministic Core projection, migrated registry and quality readers, legacy dual-read/Core single-write tests, and a dropped-zero migration gate all pass together.

Never restore a legacy YAML marker, 13/15/16-section output, placeholder sentinels, or a second authoritative Markdown dialect as a shortcut.

When Phase 2 comes back, it needs a Step 0

The Tier 1 discovery this skill used to run had three moves — guess <brand>.design, search site:<domain>, search GitHub — and all three are name-guessing or search. None asks whether the host publishes its own documentation index.

That gap has now been measured four times. Toss publishes an llms.txt with 258 links on a developer portal no name-guess reaches, and our reference documented 2 of the 11 components it lists. Pega's design.pega.com was cited correctly from the first pass and its index was never opened: 42 published components, 0 documented. KRDS publishes 55 and we cite 21. Of ten hosts with an official design-system surface, five publish a machine-readable index and none had been read.

The rule and its two traps are written up in docs/SOURCE_INDEX_BLINDNESS_2026-09-17.md (2026-09-19 section). The short version, because it is the part that is easy to get wrong:

  • A file that passes a content check is not necessarily an index. Pega's llms.txt is real, well-formed, text/plain, correctly sized — and Yoast generated it from five SEO pages. It lists zero components out of the 66 its own sitemap publishes.
  • A large index is not a component index. Kakao's sitemap is real and has 316 URLs, all of them API documentation.

Fold Step 0 in before restoring the three name-guessing moves, not after.

Write path: DESIGN.md v3 (2026-10-14)

The catalog is DESIGN.md v3 (spec/design-md-v3.md): every web/references/<id>/ holds a Google-frontmatter DESIGN.md and a reference.json sidecar. Any write this gate permits (an evidence refresh on an existing reference) goes through one path: read readReferenceSource(dir).markdown (the legacy-shaped source), edit it, and write it with scripts/design-md-v3/writer.mjs writeReference(dir, text), which converts, checks spec §10 conformance with 0 Google lint errors, and writes the file, the sidecar and the design-md/ mirror. web/scripts/assemble-reference.mjs emits v3 the same way. The OmD 0.1 text-patching tools (apply-state-values, apply-measured-palette, apply-primary-tasks, migrate-frontmatters, retire-template-motion) refuse v3 files until they are ported. Creating a new reference stays refused until the owner lifts this gate.

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.