product-selection

name

product-selection

description

Find, validate, and compare home renovation products (fixtures, hardware, appliances, finishes) with confirmed working image URLs and product links, plus community review and longevity research per candidate. Handles retailer bot-protection by routing image sourcing through accessible CDNs. Runs a structured discovery interview first to narrow requirements before searching. Output is a comparison table ready to paste into a wiki page or shareable email. ALWAYS trigger this skill when the user asks to find, research, compare, or select any physical product for the 711 N 60th remodel or any home improvement project — even if they don't say 'product selection' explicitly.

product-selection — Home Renovation Product Research & Validation

What This Skill Does

Given a product category and requirements, this skill:

1. Interviews the user to precisely define requirements before searching (avoids wasted research)
2. Searches for 2–4 options across price/quality tiers
3. Digs into reviews and longevity for each candidate — not just spec/price
4. Sources working image URLs from accessible CDNs (not guessed — validated)
5. Validates all product page links return HTTP 200 before including them
6. Formats a comparison table for wiki page or stakeholder email
7. Records the selection decision back to the relevant wiki page

Stages & Gates

This skill runs as seven discrete stages (0-6 below), each with an entry gate (what must be true to start) and an exit gate (what must be true to advance). Don't skip a gate to save time — a failed gate is a normal, expected stopping point, not a problem to route around.

Track progress with the task tool (if available). At the start of Phase 0, create one task per stage if a task tool is available — Claude Code: TaskCreate/TaskUpdate; Gemini CLI/Antigravity: write_todos. Mark each stage in_progress when it starts and completed when its exit gate passes. If no task tool is available, skip this — it's a convenience for resuming, not a requirement.

---

User Shopping Preferences

Preferred retailers — always prioritize links from these when available:

• Lowe's (lowes.com) — first preference
• Amazon (amazon.com) — second preference

When candidates are sold at Lowe's or Amazon, lead with those links in the comparison output. If a product is not available at either, note that explicitly and fall back to other retailers. Both domains block automated WebFetch (see Phase 3 domain table) — use their URLs as human-clickable links only, never for image sourcing or URL validation.

---

Phase 0 — Discovery Interview

Entry gate: none — this is the first stage. Mark its task in_progress if a task tool is available.

Run this phase before any searching. Use AskUserQuestion for each question. Do not batch — ask one at a time, wait for the answer, then adapt follow-up questions based on responses. Do not propose solutions during the interview.

HARD GATE: Do not run any WebSearch or WebFetch until Phase 0 is complete.

Question 1 — Product Category

header: "Product type"
question: "What type of product are you selecting?"
options:
  - "Plumbing fixture (faucet, sink, shower, toilet)"
  - "Door/cabinet hardware (knobs, levers, pulls, hinges)"
  - "Appliance (range, refrigerator, hood, dishwasher)"
  - "Finish material (tile, countertop, flooring, wallcovering)"

(If "Other" selected, ask for a one-line description before proceeding)

Question 2 — Finish / Material Constraint

header: "Finish"
question: "What finish or material is required?"
options:
  - "Must match an existing item — I'll specify (click Other)"
  - "Polished chrome"
  - "Brushed/satin nickel"
  - "Matte black"

Key follow-up: If "must match an existing item" — ask which item and its exact model/SKU. Finish match is the #1 cause of selection mistakes.

Question 3 — Functional Constraint

header: "Function"
question: "Are there any hard functional requirements?"
options:
  - "No hard constraints beyond category and finish"
  - "Physical constraint (hole spacing, size, rough-in) — I'll specify"
  - "Occupant constraint (e.g., child/pet-proof, ADA, elderly user)"
  - "Code or permit requirement — I'll specify"

Examples of occupant constraints: Lever handles can be opened by dogs or young children — knobs required. ADA requires lever handles (contradiction — must flag if both apply).

Question 4 — Budget

header: "Budget tier"
question: "What price tier should the options span?"
options:
  - "Budget + mid-range (show both, recommend mid)"
  - "Mid-range + luxury (show both, recommend mid)"
  - "Full range — budget through luxury"
  - "I have a specific per-unit budget (click Other)"

For hardware: always ask for quantity to compute total cost — $150/door sounds reasonable; $900 for 6 doors may change the decision.

Question 5 — Visual Coordination

header: "Coordination"
question: "What existing items does this need to visually coordinate with?"
options:
  - "Other items in the same room — I'll list them (click Other)"
  - "An existing product already specified for this project"
  - "No specific coordination requirement"
  - "I'll describe the overall design direction (click Other)"

Record all coordination constraints — they narrow options significantly and prevent mismatches.

Question 6 — Output Format

header: "Output use"
question: "How will you use the comparison?"
options:
  - "Email to stakeholder for approval (need inline images + links)"
  - "Update the project wiki page"
  - "Both — email and wiki"
  - "Just for my own reference"

This determines whether to run full image validation (email requires confirmed image URLs) or a lighter wiki-only format.

Exit gate — Pre-search Checklist

Before proceeding, confirm:

• Product category is specific (not "faucet" — "single-hole bar faucet, gooseneck, 1.8 GPM")
• Finish is locked and any "must match" item is identified by model number
• Functional constraints are captured (especially occupant constraints like pet/child)
• Quantity is known if this is hardware (knobs, hinges, pulls, etc.)
• Visual coordination notes are recorded

Do not proceed to Phase 1 until every box is checked — ask a follow-up question instead of guessing. Mark Phase 0's task completed if a task tool is available.

---

Phase 1 — Search Strategy

Entry gate: Phase 0's checklist passed. Mark this stage's task in_progress if a task tool is available.

Run 2–3 parallel WebSearch calls based on requirements gathered in Phase 0. Follow the meta-research-workflow skill's search discipline (rate limits, search caps, full URL/title/date documentation for every result).

Query Templates

"[specific product name] [finish] site:lowes.com"              ← preferred retailer
"[specific product name] [finish] site:amazon.com"              ← preferred retailer
"[specific product name] [finish] [brand if constrained] comparison options"
"[specific product name] [finish] site:[accessible-brand].com"
"[specific product name] [finish] [brand] site:pinterest.com"   ← for image sourcing

Identify Candidates

Find 2–4 candidates across price tiers. For each, record:

• Product name + model number/SKU (both manufacturer and retailer SKUs if different)
• Brand
• Approximate price (note source and date — prices drift)
• Retailer URLs (even blocked domains — they're for human-clickable links, not WebFetch)
• Any confirmed constraints it satisfies or violates

Exit gate: 2-4 candidates identified across at least 2 price tiers, each with the fields above recorded. Mark Phase 1's task completed if a task tool is available.

---

Phase 2 — Review & Longevity Research

Entry gate: Phase 1's candidate list exists. Mark this stage's task in_progress if a task tool is available.

For each candidate, apply the review-longevity-research skill: search community/long-term-use sources, weigh sponsored vs. organic signal, and check longevity proxies (warranty, repairability, construction). Produce the skill's ### Review & Longevity — <Product Name> block for every candidate.

Exit gate: every candidate has a Review & Longevity block — either real findings, or an explicit "no long-term data available" note if the product is too new. Never skip a candidate silently. Mark this stage's task completed if a task tool is available.

---

Phase 3 — Image Sourcing

Entry gate: Phase 2's Review & Longevity blocks exist for every candidate. Mark this stage's task in_progress if a task tool is available.

Fetch Tool Hierarchy

Use tools in this order when a domain blocks WebFetch:

1. WebFetch — try first; fast and sufficient for accessible domains
2. mcp__read-website-fast__read_website — try when WebFetch returns 403/429; often bypasses bot protection that blocks WebFetch. Save to /tmp if page is large.
3. Chrome MCP (mcp__claude-in-chrome__navigate + mcp__claude-in-chrome__get_page_text) — last resort for pages that require JavaScript rendering or are behind login walls. Use when read-website-fast also fails. Load tools via ToolSearch before calling.

Never give up on a preferred retailer (Lowe's, Amazon) without trying all three tools.

Critical Domain Routing

Try mcp__read-website-fast__read_website before giving up — may bypass bot protection:

Domain	WebFetch result	Fallback
lowes.com	403 Forbidden — preferred human link	try read-website-fast, then Chrome MCP
amazon.com	500 / robot challenge — preferred human link	try read-website-fast, then Chrome MCP
homedepot.com	403 Forbidden	try read-website-fast
wayfair.com	429 / 403	try read-website-fast
walmart.com	Robot challenge	try read-website-fast
signaturehardware.com	403 Forbidden	try read-website-fast
ferguson.com	Akamai bot	try read-website-fast
fergusonhome.com	403 Forbidden	try read-website-fast
ebay.com	Timeout	try read-website-fast
faucet.com	404 (stale URL patterns)	re-search for updated URL
faucetdirect.com	404 (stale URL patterns)	re-search for updated URL
build.com → fergusonhome.com	403 Forbidden	try read-website-fast

Only mark a domain ⚠ browser only after all three fetch methods have failed.

ACCESSIBLE — Use for images and validation:

Domain	Notes
i.pinimg.com	Pinterest CDN — most reliable fallback for any product
emtek.com	Full pages + CDN images
baldwinhardware.com	Product pages; images at images.baldwinhardware.com
images.baldwinhardware.com	Scene7 CDN — use ?fmt=jpeg&wid=600
kohler.com	Accessible
deltafaucet.com	Accessible
moen.com	Accessible
schlage.com	Accessible
kwikset.com	Accessible
houseofrohl.com	Official Rohl site — accessible; images on BigCommerce CDN
cdn11.bigcommerce.com	BigCommerce CDN — usually accessible
Pinterest pin pages (pinterest.com/pin/...)	Fetch og:image to get pinimg.com URL

Image Sourcing Priority Order

For each candidate product:

1. Official brand website CDN (best quality, most stable)

• Fetch the brand's own product page (only if domain is on ACCESSIBLE list)
• Extract og:image or main img src
• Emtek CDN pattern: https://www.emtek.com/media/salsify/images/[size]/[hash]-[date]-[ProductName]_EM_KO.jpg
• Baldwin CDN: https://images.baldwinhardware.com/is/image/Baldwin/[sku]?fmt=jpeg&wid=600

2. Pinterest pins (reliable fallback for any brand)

• Search: site:pinterest.com "[exact product name] [finish]"
• Prefer pins with specific product name AND finish in the title (e.g., "amberley-singlehole-bar-faucet-chrome-in-2024")
• UK/regional Pinterest pins often have cleaner titles: try uk.pinterest.com
• Fetch the pin page: WebFetch(url, "Find the og:image meta tag content URL. Return the full pinimg.com URL and confirm the product name and finish shown.")
• Validate the returned i.pinimg.com URL

3. Accessible third-party CDNs

• houseofrohl.com products use BigCommerce CDN (cdn11.bigcommerce.com)
• Some specialty retailers (patioliving.com, focalpointhardware.com) have accessible CDNs

If no image found after all three attempts: Note "image available at [brand URL] — opens in browser" in the comparison. Never fabricate or guess an image URL.

Exit gate: every candidate has either a sourced image or an explicit "opens in browser" note — never a silently missing image. Mark this stage's task completed if a task tool is available.

---

Phase 4 — URL Validation

Entry gate: Phase 3's image sourcing pass is done for every candidate. Mark this stage's task in_progress if a task tool is available.

Validate every URL before including it — no exceptions.

Image URL Validation

Try tools in order until one succeeds:

1. WebFetch(url=image_url, prompt="Is this a valid accessible image? Return HTTP status and approximate file size in KB.")
2. mcp__read-website-fast__read_website(url=image_url) — if WebFetch blocked
3. Chrome MCP navigate + screenshot — if both above fail and domain is a preferred retailer

Interpreting results (WebFetch/read-website-fast can't render binary images — binary saves are valid):

• Valid signal: Binary JPEG/PNG file saved, size > 5KB → confirmed working image
• Invalid: HTTP 403 or 404 on all three tools → discard, find alternative
• Ambiguous: File saved but < 3KB → likely a placeholder or error icon → discard

Product Page Validation

Try tools in order:

1. WebFetch(url=product_url, prompt="Return HTTP status and page title.")
2. mcp__read-website-fast__read_website(url=product_url) — if WebFetch blocked
3. Chrome MCP navigate — if both above fail; navigate and read page text

• Success + matching page title → confirmed ✓
• HTTP 404 on all tools → URL is stale; search for updated URL on the same domain
• All tools blocked → note ⚠ browser only; URL still usable as human link

Validation Status Tags

Use these in the final output:

• ✓ confirmed — any fetch tool returned 200 + matching content
• ✓ image confirmed — image URL returns JPEG/PNG > 5KB via any tool
• ⚠ browser only — all three fetch methods blocked; product link confirmed accessible in user's browser
• ❌ broken — 404 confirmed across all tools with no working alternative found

Exit gate: every product page URL and image URL carries one of the four status tags above — nothing left unvalidated. Mark this stage's task completed if a task tool is available.

---

Phase 5 — Format Output

Entry gate: Phase 4's validation pass is complete for every candidate. Mark this stage's task in_progress if a task tool is available.

Comparison Table (wiki format)

## [Category] Options

**Requirements**: [finish] · [functional constraint] · [coordination note]

| Option | Brand/Model | Price | Image | Link | Status |
|--------|-------------|-------|-------|------|--------|
| ⭐ A (Recommended) | Brand Model SKU | ~$X | [confirmed image URL] | [product link] | ✓ confirmed |
| B | ... | ~$X | [confirmed image URL] | [product link] | ✓ confirmed |
| C | ... | ~$X | N/A | [product link] | ⚠ browser only |

**Recommendation**: [1–2 sentences — finish match, value, design rationale, and any review/longevity signal that influenced the pick]
**Total cost** (if hardware): [unit price × quantity] = [range]

<!-- Append each candidate's Review & Longevity block from Phase 2 below the table -->

Email Format (stakeholder-shareable)

When output is for email, use inline images with Markdown:

### ⭐ Option A — [Brand Model] · ~$[price] · RECOMMENDED

![Alt description](https://confirmed-image-url.jpg)
👉 [View on [Site]](https://product-page-url)

| | |
|---|---|
| **Brand** | ... |
| **Model** | ... |
| **Finish match** | [specific note] |
| **Price** | ~$X |

Email format rules:

• Every option must have either a confirmed inline image OR an explicit note explaining why the image isn't embeddable
• All product links must be validated (200 or noted as "browser only")
• Total cost at bottom if hardware
• Summary table of recommendations at the end

Exit gate: the comparison table (and email format, if requested) is complete, every option has a status tag, and the Review & Longevity blocks are attached. Mark this stage's task completed if a task tool is available. Present the comparison to the user and wait for a selection before Phase 6.

---

Phase 6 — Record Decision

Entry gate: the user has selected an option from Phase 5's comparison. Mark this stage's task in_progress if a task tool is available.

Apply the knowledge-synthesis skill's Decision Write-Back Pattern to record the choice on the relevant wiki page in logseq/pages/. Product-selection specifics on top of the generic pattern:

1. Update the options table to mark the selected item
2. Move the item from "Outstanding Items — Owner's Side" to "Decisions Made / Locked"
3. Status progression: [ ] Pending approval → [x] Approved → [ ] Ordered
4. Link the chosen option to its Product & Retailer Zettel (per the Decision Write-Back Pattern) if one was created — carry over the Review & Longevity findings from Phase 2 onto that page rather than losing them once the decision is made

## Decisions Made / Locked
- **[Category]**: [[Brand Model SKU]] — [finish] — [rationale, e.g., "matches Levi faucet finish, same brand, stronger long-term reliability record"] — [x] Approved / [ ] Ordered

Exit gate: the wiki page reflects the decision (moved out of outstanding, status set, linked to the product zettel if one exists). Mark this stage's task completed if a task tool is available — this is the last stage.

---

Known Product Category Sources

Plumbing Fixtures

Brand	Website	Image access	Notes
Signature Hardware	signaturehardware.com	❌ 403	Use Pinterest pins for images
Rohl / Perrin & Rowe	houseofrohl.com	✓	Images on BigCommerce CDN
Kohler	kohler.com	✓	CDN at content.kohler.com
Delta / Brizo	deltafaucet.com	✓
Moen	moen.com	✓
Hansgrohe	hansgrohe-usa.com	✓

Door & Cabinet Hardware

Brand	Website	Image access	Notes
Emtek	emtek.com	✓	Images at /media/salsify/images/
Baldwin Reserve	baldwinhardware.com	✓	Scene7 CDN at images.baldwinhardware.com
Schlage	schlage.com	✓
Kwikset	kwikset.com	✓
Top Knobs	topknobs.com	check

Appliances

Brand	Notes
Viking	vikingrange.com
Bosch	bosch-home.com/en-us/
Zephyr	zephyronline.com
Wolf	subzero-wolf.com

Tile & Stone

Brand	Notes
Zia Tile	ziatile.com
Fireclay	fireclaytile.com
Ann Sacks	annsacks.com
Caesarstone	caesarstoneus.com

---

Common Failure Modes & Recovery

Failure	Cause	Fix
Image URL returns 403/404	Retailer CDN blocking or stale URL	Try Pinterest pin for same product
Image returns < 3KB	Placeholder/icon served instead	URL is wrong; search for correct CDN path
Product page returns 404	Retailer restructured URLs	Re-search site:[domain] "[product name]"
No image found on any source	Product only sold on blocked retailer	Note "images at [URL] — opens in browser"; provide description instead
Pinterest pin shows wrong finish	Multiple finishes share pins	Search for finish-specific pin (e.g., add "chrome" or "polished" to search)
Price seems wrong	Prices drift; older search results	Cross-reference 2 sources; note "approximate" and date

---

Quality Checklist Before Delivering Output

• All image URLs return JPEG/PNG > 5KB (or explicitly noted as unavailable)
• All product page links return 200 or noted as "browser only"
• Finish described exactly matches requirement (e.g., "polished chrome" ≠ "chrome" ≠ "brushed chrome")
• Model numbers confirmed against ≥ 2 sources
• Prices marked "approximate" with basis noted
• Total cost calculated if hardware (unit × quantity)
• At least 2 price tiers represented
• Recommendation includes finish-match and design rationale
• Functional constraints satisfied (checked against Phase 0 answers)
• Occupant constraints noted (dog-proof, ADA, child-safe, etc.)
• Every candidate has a Review & Longevity block (Phase 2) — real findings or an explicit "no long-term data" note
• Every non-obvious claim (price, review, longevity, failure mode) has a source URL per the meta-research-workflow skill

---

Related Skills

• [[design-review]] — Verify selected products against CD set specs before ordering
• [[knowledge-synthesis]] — Synthesize product research into permanent wiki notes; supplies the Product & Retailer Zettel Template (Phase 2/3) and Decision Write-Back Pattern (Phase 6)
• [[review-longevity-research]] — Phase 2's method for community-review and longevity digging
• [[meta-research-workflow]] — Search methodology and sourcing discipline used throughout
• [[handy:plan]] — Plan installation of selected products