- Home
- Skills
- Branding & Design
- sanstudio Design Chain
sanstudio Design Chain
A design pipeline for your agent that researches before it designs, offers three real directions, reviews its own mockup with fresh eyes, and waits for your approval at every gate.
Free
Works with the AI Agents you already use
sanstudio Design Chain
Example session with this skill installed
lane: standard
Our budgeting app's transaction list feels cluttered. Users say they can't find last week's coffee. Screenshot attached. Keep the tab bar. iOS first.
- Read your context and instructions
- Compiled the sanstudio design chain
Illustrative session, abridged. The chain stops after every gate; each block below is one stop, and the user's reply is shown in italics.
DESIGN CHAIN READY
contract: none product: none taste: user taste file
lane: standard (redesign of one screen, approach open)
G0 Research
| Id | Finding | Confidence |
|---|---|---|
| C1 | Users report trouble finding one past payment | problem: supported (2 quotes); cause: plausible |
| C2 | The list is flat reverse-chronological with no grouping | observed in screenshot |
| Gap | Are people searching for one payment or scanning a week? | decides the direction |
G0 done. Reply APPROVE, ITERATE, BACK or STOP. APPROVE
G1 Context lock
- Visitor mode: Operate. Invention scope: surface (existing app look kept).
- Category default to refuse: a flat dated list with category color chips.
- Tokens extracted from the screenshot into the design contract: real hex values, SF Pro, 16px gutters.
- One question: is search already in the app? "No, add it if it helps." APPROVE
G2 Three directions
| Direction | Answers | Differs on | |
|---|---|---|---|
| A | Day headers with a fact row per day (count, total) | C1 | information architecture |
| B | Grouped by merchant, last visit on top | C1, gap | information architecture |
| C | Search-first, list as results | gap | interaction |
Recommended: A. The familiar version is offered as a quiet option. "B, with A's day headers." The direction contract (idea, look, journey, first screen, signature move, done when) is written to the brief.
G3 Hi-fi and review
- A hi-fi HTML mockup in a phone frame, real merchants and amounts, no dots in chips, no bordered cards.
- Anti-slop pass: clean. Captures taken at 390 x 844.
- Independent review: heuristics 29/40; persona walk as a one-thumb commuter and a screen-reader user;
one P1: no empty state for a merchant with no visits this month. Fixed before the user saw the screen. Disposition: ship.
- Figma export JSON written to the session folder.
ITERATE quieter tones down the merchant category colors without touching layout, then the review re-runs on the changed area.
Next steps: APPROVE to finish, or take it to G4 for a production package (all states, components, handoff).
Connects securely to your tools. The creator never sees your data.
About this skill
A design pipeline for your agent that researches before it designs, offers three real directions, reviews its own mockup with fresh eyes, and waits for your approval at every gate.
What it does. Six skills that turn a design request into artifacts you can keep: a research report, a session DESIGN.md with tokens and rules, three directions that differ in structure (not color), a hi-fi HTML mockup with real content, a review with heuristic scores and P0 to P3 issues, Figma JSON, and on the major lane a production package with states, components and a readiness scorecard.
How it works. design-chain picks a lane (fast, standard, major), a visitor mode (Persuade, Operate, Read, Experience) and an invention scope, then runs the gates in order. Research always comes first. Directions refuse the category's default page. Gate 3 is checked against 50+ craft rules (no dots in chips, no gradient text, generous spacing, chosen fonts) and then reviewed from rendered captures by a context that did not build it. After every gate you reply APPROVE, ITERATE, BACK or STOP; refine verbs such as bolder, quieter and distill give scoped passes.
| Gate | Skill | Output |
|---|---|---|
| entry | design-chain | lane, mode, gate control |
| G0 | research-synthesis | research report with findings and gaps |
| G1 | context-lock | session DESIGN.md (tokens and rules) |
| G2 | direction-lock | three directions and a direction contract |
| G3 | design-lock | hi-fi HTML, captures, review, Figma JSON |
| G4 | production-lock | states, components, handoff, readiness scorecard |
Where it stops. It designs; it does not write your product code or merge anything. Mockups are proposals, not shipped UI. It uses the tools your agent already has (search, browser, Figma MCP) and records any missing tool as an evidence gap instead of faking it. Your brand stays yours: a TASTE.md, PRODUCT.md or .sanstudio/ contract in your project overrides its defaults.
Credits. Original pipeline and rules, with a few ideas adapted in new wording from impeccable (Apache 2.0), design.md (Apache 2.0), hallmark (MIT) and taste-skill (MIT). Full credits and licence texts ship in NOTICE.md and licenses/.
Designs, does not code: mockups are HTML proposals, never your product code, and it never edits your repository. It stops at every gate and waits for your reply, so it is not a one-shot generator. Output quality depends on the tools your agent has; a missing tool (image search, browser, subagents) is recorded as an evidence gap rather than faked. Gate 4 produces a buildable design spec with an honest readiness scorecard, not shippable code.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 19 days ago
- Free with an Agensi account
Permissions
Read and write files: the chain reads your TASTE.md, PRODUCT.md and .sanstudio/ folder, and writes its artifacts (research report, DESIGN.md, HTML mockups, review, Figma JSON) to ./sanstudio-output/ in your project. Network: only through your agent's own web and image search during research and image sourcing; the pack ships no scripts and makes no network calls itself. Browser: renders the mockup to capture screenshots for the review. It never writes your product code.