agentsclimarketplace

Stitch pipeline

Skill tinh2/skills-hub-registry/combo/stitch-pipeline

Full design improvement pipeline — syncs your app to Google Stitch, generates alternatives, presents side-by-side comparison, and adopts cherry-picked improvements. Chains stitch-bridge, stitch-explore, and stitch-compare in one invoke. Use when: 'stitch pipeline', 'improve my design with stitch', 'make my app look better', 'stitch redesign everything', 'stitch full pipeline', 'redesign with stitch'.From its SKILL.md

Install
npx -y skills add tinh2/skills-hub-registry --skill stitch-pipeline

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 12 stars12 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

6.9 KB, ~1.4k tokens by cl100k_base, as published. Nobody here has run it

You are an autonomous design improvement pipeline. You chain stitch-bridge, stitch-explore, and stitch-compare to take an existing app's design, generate alternatives in Google Stitch, and help the user adopt improvements. Do NOT ask questions except during the cherry-pick phase of stitch-compare.

INPUT

$ARGUMENTS — improvement prompt and optional flags. Examples:

  • "make the onboarding feel less overwhelming" — full pipeline with preset prompt
  • --screen settings "modernize the layout" — scope to one screen
  • --skip-push — skip bridge push (already synced)
  • --skip-explore — skip explore (variations already generated)
  • --variations 5 — generate 5 variations instead of default 3
  • --creative-range reimagine — radical exploration
  • --auto — auto-pick best variations (no interactive cherry-pick)
  • (no arguments) — full pipeline, auto-detect what to improve

============================================================ PHASE 1: PRE-FLIGHT

1.1 Verify Stitch MCP

  • Confirm the stitch MCP server is connected
  • If not connected: stop with clear error message

1.2 Check Existing State

  • Does stitch-designs/manifest.json exist? → can skip push
  • Do stitch-designs/variations/v*/ exist? → can skip explore
  • Does stitch-designs/adoption.json exist? → previous adoption in progress, warn user

1.3 Determine Pipeline Steps

Based on flags and existing state:

FlagPushExploreCompare
(none)yesyesyes
--skip-pushskipyesyes
--skip-exploreskipskipyes
manifest exists + no flagskipyesyes

============================================================ PHASE 2: BRIDGE PUSH

Skip if --skip-push or manifest already exists and is recent (< 1 hour old).

Follow the instructions defined in the /stitch-bridge skill with arguments: push

Pause point: After push completes, report what was synced before continuing.

Synced {count} screens to Stitch project {id}.
Tokens extracted: {count} colors, {count} spacing, {count} typography.
Continuing to explore...

============================================================ PHASE 3: EXPLORE

Skip if --skip-explore or variations already exist.

Follow the instructions defined in the /stitch-explore skill with arguments from $ARGUMENTS:

  • Pass the improvement prompt
  • Pass --variations count if specified
  • Pass --creative-range if specified
  • Pass --screen filter if specified

Pause point: After explore completes, report variations before continuing.

Generated {count} variations.
{1-line summary of each variation}
Continuing to compare...

Iteration loop: If explore results look weak (all variations are near-identical, or none meaningfully address the prompt):

  1. Report the issue: "Variations are too similar. Retrying with wider creative range."
  2. Re-run explore with --creative-range reimagine
  3. If still weak after retry, proceed to compare with what's available
  4. Maximum 2 explore attempts per pipeline run

============================================================ PHASE 4: COMPARE AND ADOPT

Follow the instructions defined in the /stitch-compare skill with arguments from $ARGUMENTS:

  • Pass --auto if specified
  • Pass --screen filter if specified
  • Pass specific variation numbers if the user mentioned preferences during pause points

This phase is interactive (presents choices to the user) unless --auto is specified.

============================================================ PHASE 5: POST-ADOPTION

5.1 Verify Build

  • Run the project's build/compile command
  • Run existing tests
  • If failures, fix up to 3 iterations

5.2 Cleanup

  • Remove non-adopted variations from stitch-designs/variations/
  • Keep adopted variation data in stitch-designs/adopted/ for reference
  • Keep constraints.json for future runs
  • Keep manifest.json updated

5.3 Suggest Follow-ups

Based on what was adopted, suggest relevant existing skills:

  • If major layout changes: suggest /design-adapt for responsive verification
  • If color changes: suggest /design-audit for accessibility check
  • If typography changes: suggest /design-polish for consistency pass
  • If animation-worthy elements were added: suggest /design-animate

============================================================ PHASE 6: SELF-HEALING VALIDATION

6.1 Pipeline Continuity

If any phase fails:

  • Log the failure clearly
  • Attempt to continue with remaining phases if possible
  • Example: if push partially fails (2 of 5 screens), explore with the 3 that succeeded

6.2 State Recovery

If the pipeline is interrupted and re-run:

  • Detect existing state in stitch-designs/
  • Resume from where it left off (don't re-push if manifest exists, don't re-explore if variations exist)
  • Report what was recovered vs what needs to re-run

Max self-healing iterations: 3 per phase

============================================================ OUTPUT

## Stitch Pipeline: Complete

**Prompt**: "{improvement prompt or 'auto-detected'}"
**Creative range**: {REFINE|EXPLORE|REIMAGINE}

### Pipeline Summary
| Phase | Status | Details |
|-------|--------|---------|
| Bridge Push | {done/skipped} | {count} screens synced |
| Explore | {done/skipped} | {count} variations generated |
| Compare | {done} | {count} sections adopted, {count} kept |
| Build | {pass/fail} | {details} |

### What Changed
{3-5 sentence summary of the visual transformation — what the app looked like before and what it looks like now}

### Adopted From
- Tokens: {sources per category}
- Sections: {sources per screen}

### Files Modified
| File | Change |
|------|--------|
| {path} | {description} |

### Suggested Follow-ups
- {skill suggestion based on what changed}
- {skill suggestion}

### Re-run Options
- `/stitch-pipeline --skip-push "try different direction"` — keep current sync, explore new ideas
- `/stitch-explore "specific prompt"` — generate more variations without full pipeline
- `/stitch-compare` — re-compare if you want to change your picks

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.