agentleFS
AcKeskinAcKeskin/contexture/skills/design-proposal/SKILL.md

design-proposal

Skill2 starsChanged 7 days agoMITView on GitHub

Use when the user asks for a page, artifact or set of pages that presents a design, proposal, architecture or research outcome to a team for discussion, or asks for a presentation or playground copy of such a page.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

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

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈8

    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:

    Read what the user points at: a prior artifact, a doc, a ticket, a conversation.

    line 18
  2. Distinctive:

    Ask nothing that the source, the code or the repo already answers.

    line 20
  3. Distinctive:

    Check every named identifier (function, type, extension, endpoint, flag, table) against its source of truth: headers, schemas, specs, the …

    line 25
  4. Distinctive:

    Search the project for its own brand: theme or color token files, a design-system package, a logo SVG's fills, an app style sheet.

    line 33
  5. Distinctive:

    When the project has none, ask the user once which brand to follow, or use a neutral palette.

    line 34
  6. Very specific:

    Start from assets/page-shell.html: tabbed pages, topic blocks, signature highlighting and diagram classes, all driven by color tokens.

    line 35
  7. Very specific:

    Every area tab gets at least one diagram of its mechanism: a sequence for a protocol, a flow for data, a mapping band for translations, a …

    line 40
  8. Distinctive:

    Write interfaces as real signatures or schemas, in the notation the team will implement.

    line 41
  9. Very specific:

    Run the humanize skill on the prose when it is installed.

    line 45
  10. Distinctive:

    When a comment asks for a change: re-read the part, change it, then update every other page that mentions the same choice.

    line 52
  11. Distinctive:

    When the user asks a question about the design without asking for an edit, answer in chat and leave the page alone.

    line 53

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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

design-proposal

Builds a page a team discusses together. The page shows what is planned and, where they exist, the other ways it could go with their limitations. Domain-agnostic: an API family, a service split, a data model, a UI flow or a process change all get the same treatment.

The page, in one paragraph

The readers already know the domain and why the work exists. Each page covers one area of the design. It opens with a one-line lede and a diagram of the mechanism, then the interfaces in the design's own notation, then one topic block per open choice. A topic block is Plan (the choice, one or two plain sentences, with the reason only when it is not obvious), followed by Other ways: a two-column table of each alternative and its limitation. A known cost of the plan itself goes in one "The catch:" line under the plan. That is the whole vocabulary of the page. Full recipe and tripwires: references/page-recipe.md.

Procedure

1. Gather the source

  • Read what the user points at: a prior artifact, a doc, a ticket, a conversation. When the source is an artifact or page, read all of it.
  • Extract only forward-looking content: the plan, the choices, the alternatives. Leave behind how the work got here.
  • Ask nothing that the source, the code or the repo already answers.

2. Research before writing

  • Look for work already in flight on the topic and shape the plan around it: recent branches, open PRs, design docs, tickets. Delegate this sweep to a search agent when it spans many places. A plan that ignores the team's current modules gets rejected in the room.
  • Check every named identifier (function, type, extension, endpoint, flag, table) against its source of truth: headers, schemas, specs, the codebase. Report in chat any name you cannot confirm; do not print it on the page as fact.

3. Write the argument

Before any page, list each tab and each topic block as one line: the claim it makes and the proof it carries (a diagram, a signature, a number, a source). Delete any block with no claim. The claims decide which tabs exist and their order. This is what stops a stock sequence of sections that each say a little.

4. Pick the look

  • Search the project for its own brand: theme or color token files, a design-system package, a logo SVG's fills, an app style sheet. Use those colors and the brand font, provided it is on Google Fonts.
  • When the project has none, ask the user once which brand to follow, or use a neutral palette. Never carry a brand over from another project.
  • Start from assets/page-shell.html: tabbed pages, topic blocks, signature highlighting and diagram classes, all driven by color tokens. Fill its token slots and follow the artifact-design skill's page contract (load it before writing).

5. Write the pages

  • One tab per area. The first tab is an overview: one diagram of the whole, and a table of the parts (name, side or owner, where it attaches, what it does).
  • Every area tab gets at least one diagram of its mechanism: a sequence for a protocol, a flow for data, a mapping band for translations, a timeline for lifetimes, side by side for a comparison. Use the artifact-diagramming skill.
  • Write interfaces as real signatures or schemas, in the notation the team will implement.

6. Copy pass

  • Run the humanize skill on the prose when it is installed. Otherwise apply the copy rules in references/page-recipe.md.
  • Then run the tripwires in references/page-recipe.md against the file. Every hit is removed or rewritten before publishing.

7. Publish and iterate

  • Gate. Publish only when all of these hold on the final file: every block on the argument list made it onto a page; every identifier is confirmed or reported in chat; the tripwires ran with zero hits left; you opened the page at desktop and phone width and read every tab. A step you skipped is named in your message, never implied.
  • Publish as an artifact and give the link. Tell the user it is private until they share it.
  • When a comment asks for a change: re-read the part, change it, then update every other page that mentions the same choice. A choice that flips moves the old plan into Other ways with its limitation. Reply in the thread and resolve it.
  • When the user asks a question about the design without asking for an edit, answer in chat and leave the page alone.

8. Playground copy (on request)

When the user wants to present or discuss on the page itself, build a second artifact from the same file with the annotation layer: inline assets/annotation-layer.css into the page style and assets/annotation-layer.js before </body>, append " playground" to the title, and publish it under a new file name so the original stays clean. The layer gives pen, marker, text, sticky notes, eraser, laser, undo, hide, full screen and arrow keys between tabs. Drawings stay in the viewer's browser; offer shared storage only if the team needs to draw together.

Relationship to other skills

  • artifact-design, artifact-diagramming: the page contract and diagram mechanics this skill builds on.
  • humanize: the copy pass.
  • human-view: projects an LLM planning artefact for one approver. This skill builds a team-facing discussion page.

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.