Yaz Skill

Yaz is a connector to the organization design and product image catalog system. It knows about org designs (logos, artwork, etc.), their approval status, and which products carry those designs — including what image files are available for each.

0. Preflight Check

Before anything else, verify the yaz MCP tools are available: mcp__yaz__search_orgs, mcp__yaz__get_org_designs, mcp__yaz__get_org_images, mcp__yaz__get_org_product_images.

If any are missing, stop and tell the human: “The Yaz MCP connector isn’t configured. Set it up first.”

1. Understand the Request

Accept input in any of these forms:

• Org name — e.g. /yaz Youth Soccer Club → look up the org and show its designs
• Design query — e.g. /yaz show approved designs for Acme → filter by approval status
• Product image query — e.g. /yaz show black polo images for Youth Inc → filter by product name and color
• Audit request — e.g. /yaz which designs for X are missing SVG files? → find gaps
• No argument — ask the human what org or design they want to explore

Parse the intent from args or conversation context. If ambiguous, make the most reasonable call and proceed — you can always refine after showing results.

2. Resolve the Org

Use mcp__yaz__search_orgs with a partial name from the request. If multiple orgs match, present them as a short numbered list and ask the human to pick. If exactly one matches, proceed without asking.

If no org is found, tell the human and stop — do not guess or substitute.

3. Route to Intent

Determine the primary intent before fetching anything beyond the org:

• Product image query (human asks about specific products, colors, or product-level images) → skip to 4c directly using the org ID — do not call get_org_designs first
• Everything else → call mcp__yaz__get_org_designs with the org ID, then route to the appropriate section below

When fetching designs, apply filters if the request implies them. Accepted filter values:

• approval_status: approved, to_review, in_progress, to_rework
• active_status: active, inactive, amazon-only

If get_org_designs returns zero results, tell the human — don’t silently render an empty table.

4. Handle the Intent

4a. Browse Designs

Show a summary table:

## Designs for {Org Name}

| ID | Name | Approval | Active |
|----|------|----------|--------|
| 12 | Primary Logo | approved | active |
| 34 | Alternate Mark | to_review | active |
...

If there are more than 15 designs, group by approval status and show counts per group first, then ask if they want to drill into a specific group.

4b. Inspect a Specific Design’s Images

When the human picks a design (or the request is for a specific design’s files), use mcp__yaz__get_org_images with the org_design_id.

Optionally filter by image_type. Accepted values: svg, base, embroidery_emb, embroidery_dst.

Show results:

## Images for Design: {design name} (ID: {id})

| Type | URL |
|------|-----|
| svg  | https://... |
| base | https://... |
| embroidery_emb | https://... |

If no images exist for a type the human asked about, call it out explicitly.

4c. Product Image Lookup

When the human wants to see what products carry a design or filter by product attributes, use mcp__yaz__get_org_product_images.

Key filter rules — and tell the human about these when a search returns empty or unexpected results:

• Use product_name for style/brand keywords (e.g. “polo”, “jersey tank”)
• Use product_color for color (e.g. “black”, “heather navy”) — SEPARATE from product_name
• Never combine style + color in a single product_name value — it will not match
• All filters are case-insensitive substring matches
• Timestamp filters are ISO 8601 UTC

Show results grouped by image URL (as the API returns them), with a list of products sharing each image:

## Product Images — {Org Name}
Filtered by: {applied filters}

### Image 1
**URL:** https://...
**Type:** base
**Design:** Primary Logo
**Products using this image:**
- Youth Polo S Black
- Youth Polo M Black
- Youth Polo L Black

### Image 2
...

If the result set is large (>20 images), summarize counts first and ask if they want the full list or a more specific filter.

If results are empty, explain why the filter may have missed (e.g. style+color combined in one field) and suggest a corrected query.

4d. Audit / Gap Analysis

When the human asks to find missing images, coverage gaps, or incomplete designs:

1. Fetch all designs for the org (optionally filtered by status)
2. Gate before fan-out: if there are more than 20 designs, tell the human the count and ask to confirm before proceeding — fetching images for each design is one tool call per design
3. For each design (within the confirmed batch), call mcp__yaz__get_org_images — run in parallel
4. Check which image types are present vs. missing
5. Report gaps:

## Image Coverage Audit — {Org Name}

| Design | SVG | Base | Embroidery EMB | Embroidery DST |
|--------|-----|------|----------------|----------------|
| Primary Logo | ✓ | ✓ | ✗ | ✗ |
| Alternate Mark | ✗ | ✓ | ✗ | ✗ |
...

### Summary
- {N} designs missing SVG
- {N} designs missing all embroidery files
- {N} designs fully covered

Flag any design with approval_status=approved that is missing files — those are the most urgent gaps.

5. Offer Follow-up Actions

After any result, offer relevant next steps:

• Filter further — apply different status, color, or product filters
• Audit coverage — run a gap analysis across all designs
• Done — nothing more needed

Don’t offer options that don’t make sense for the current result. Keep it short.