AgentStack
SKILL verified MIT Self-run

Authoring Stitch Prompts

skill-joaquimscosta-arkhe-claude-plugins-authoring-stitch-prompts · by joaquimscosta

Converts natural-language descriptions or UI spec files into optimized Google Stitch prompts. Use when creating, refining, or validating design directives for Google Stitch. Use when user says "create a Stitch prompt", "optimize this for Stitch", "convert this spec to a Stitch prompt", "write a UI prompt", or mentions Google Stitch prompt authoring. Follows Stitch best practices with short, direc…

No reviews yet
0 installs
18 views
0.0% view→install

Install

$ agentstack add skill-joaquimscosta-arkhe-claude-plugins-authoring-stitch-prompts

✓ scanned · ✓ verified — works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • Prompt-injection patterns
  • Secret / credential exfiltration
  • Dangerous shell & filesystem operations
  • Untrusted network calls
  • Known-malicious package signatures

What it can access

  • Network access No
  • Filesystem access No
  • Shell / process execution No
  • Environment & secrets No
  • Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

Are you the author of Authoring Stitch Prompts? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

Authoring Stitch Prompts

Quick Start

  1. Collect context – accept natural language, specs, or referenced files describing the screen/app.

1.5. Discover design context (optional) – check for design-intent/:

  • If exists: Extract Project Type, Design System from design-intent/memory/constitution.md
  • If not found: Scan codebase for framework hints (package.json)
  • Falls back gracefully to standalone mode
  • See [WORKFLOW.md](WORKFLOW.md#05-design-context-discovery-optional-enhancement) for details.
  1. Parse essentials – identify app type, screen focus, layout elements, and visual cues.
  2. Detect split points – analyze if input contains multiple screens or distinct intents (>2). Apply smart defaults: split if >2 screens/intents, else combine. Users can request regeneration with different approach.
  3. Filter aggressively – strip ALL non-UI concerns (backend, auth, APIs, caching, error handling, performance metrics, code-level specs). Focus EXCLUSIVELY on visual layout, components, colors, typography, spacing, and interaction patterns.
  4. Condense – rewrite into one atomic Stitch directive using "Design/Create/Add…" phrasing.
  5. Structure output – follow the Stitch prompt template (directive sentence → bullet list → 3–6 style cues → constraints). If design context was discovered, inject project-appropriate style cues. Do NOT use multi-section headings.
  6. Validate – ensure UI nouns are present, word count 6 prompts, split into prompt-v{version}-part{N}.md files (max 6 per part)
  • Copy-paste ready: Entire file works directly in Stitch interface

See [WORKFLOW.md](WORKFLOW.md#35-generate-layout-prompt-if-detected) for detailed file composition rules, [REFERENCE.md](REFERENCE.md) for Stitch best practices, and [EXAMPLES.md](EXAMPLES.md) for worked examples (Examples 14–16 cover multi-component and split scenarios).


Input Types

Accepted

  • Natural-language descriptions (single screen or short flows)
  • Markdown/YAML/JSON specs (/specs/dashboard.md)
  • Revision directives ("move KPI cards above chart", "convert to French", "change button to green")
  • References to uploaded wireframes or images
  • Language conversion requests ("switch to Spanish", "German version")
  • Structured input from /prompt command (see below)

Structured Input (from /prompt command)

When invoked via /prompt, the skill may receive pre-parsed preferences (Brief, Components, Style, Structure). Parse structured fields and apply style mappings before proceeding. See [WORKFLOW.md](WORKFLOW.md#08-parse-structured-input-if-present) for field handling and style mapping tables.

All input detail levels are valid — Stitch infers patterns from minimal descriptions. Use adjectives to convey vibe when details are sparse ("vibrant fitness app", "minimal meditation app").


Workflow Overview

High-level loop: parse → condense → format → validate. Detailed branching logic, including cue extraction and revision handling, lives in [WORKFLOW.md](WORKFLOW.md).


Output Structure

Prompts must follow the Stitch-friendly template:

  • One-sentence description of the app/screen + primary intent.
  • Bullet list (3–6 items) covering layout, components, or flows.
  • Visual style cues (palette, typography, density, tone).
  • Optional behavior/constraint reminders (responsiveness, export format).

Reference [templates/authoring-stitch-prompts-template.md](templates/authoring-stitch-prompts-template.md) for wording patterns and [templates/layout-prompt-template.md](templates/layout-prompt-template.md) for layout/foundation prompts.


Examples

Representative before/after samples (SaaS dashboard, banking app, iterative edits, spec conversions) are in [EXAMPLES.md](EXAMPLES.md). Use them to mirror tone and formatting; keep this file lean by not re-embedding the full transcripts here.


Design Context Integration

When design-intent/ exists in the project, the skill enhances style cues with project context:

  • Project Type influences tone (e.g., "enterprise-grade" for Enterprise, "friendly, approachable" for Consumer)
  • Design System names appear in style cues (e.g., "Fluent UI styling", "Material Design patterns")

The skill does NOT inject specific tokens (hex colors, spacing values)—only high-level descriptors that help Stitch generate contextually appropriate designs.

Fallback behavior: If design-intent/ is not found, the skill works standalone using default style cues.

See [WORKFLOW.md](WORKFLOW.md#05-design-context-discovery-optional-enhancement) for discovery logic and [WORKFLOW.md](WORKFLOW.md#37-inject-design-context-into-style-cues) for injection rules.


MCP Integration (Optional)

When @_davideast/stitch-mcp is configured, prompts can be sent directly to Stitch for generation after authoring.

With MCP: Author prompt -> Generate screens -> Fetch images/code Without MCP: Author prompt -> Copy to Stitch manually

After authoring, offer: "Stitch MCP is available. Generate screens now? [Yes / No / Just save prompts]"

If accepted, invoke the generating-stitch-screens skill with the authored prompt file.


Common Issues

  • Prompts too verbose – Re-run formatting with the template and trim narration. See [TROUBLESHOOTING.md](TROUBLESHOOTING.md#prompts-are-too-verbose).
  • Missing style cues – Derive palette/typography keywords from user input or prior session context before finalizing. See [TROUBLESHOOTING.md](TROUBLESHOOTING.md#missing-style-cues).
  • Multi-goal briefs – Split into multiple prompts; re-emphasize Stitch's atomic focus. See [TROUBLESHOOTING.md](TROUBLESHOOTING.md#multi-screen-prompts).

Reference Files

For advanced usage:

  • [REFERENCE.md](REFERENCE.md) — Overview of Stitch best practices
  • [EXAMPLES.md](EXAMPLES.md) — Sample transformations
  • [WORKFLOW.md](WORKFLOW.md) — Detailed processing loop
  • [TROUBLESHOOTING.md](TROUBLESHOOTING.md) — Error-handling guidance
  • [templates/authoring-stitch-prompts-template.md](templates/authoring-stitch-prompts-template.md) — Output format template

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet — be the first.

Versions

  • v0.1.0 Imported from the upstream source.