product-audit-pipeline
Skill2 starsChanged 5 months agoMITView on GitHub
Site-wide product-page (PDP) audit orchestrator for Shopify stores. Pulls ALL products via Shopify connector pagination, runs a 2-tier audit (Tier 1 metadata-only quick-scan on all products + Tier 2 full product-analyze deep-scan on the prioritized subset), ranks the portfolio by opportunity (traffic × gap severity), and escalates any open Merchant-Center feed-risk gate to the top of the queue regardless of score. Produces a ranked portfolio report (HTML/MD + per-product scores + a prioritized action queue). Use when user says "audit all products", "product portfolio audit", "which products need optimizing", "product page audit", "find weak product listings", "PDP audit pipeline", or any whole-site Shopify product-page review.
In this repository7 other files it gives its agents.
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 3 are distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 58
Verify Shopify connector is authenticated and matches the site:
- Distinctive: line 62
If mismatch, fail fast with "switch-shop required" instruction.
- Distinctive: line 137
Before ranking by opportunity, run a lightweight feed-risk check against the Phase 1+4 cache for every product (this does NOT require the …
- Very specific: line 201
If --no-gsc not set, mandate GSC pull for each (already cached from Phase 4 — reuse, don't re-pull).
- Very specific: line 249
Does NOT optimize or rewrite product copy/schema, and does NOT push any change to Shopify (audit-only) — that's product-optimize
- Very specific: line 251
Does NOT replace product-analyze for single-page deep audits — that skill remains the canonical "I want to deep-dive THIS product" tool
- Very specific: line 253
Does NOT resolve feed-risk flags itself — it surfaces and prioritizes them.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Product Audit Pipeline: Site-Wide PDP Portfolio Analyzer
Sister skill to collection-audit-pipeline but for Shopify product pages
(PDPs). Runs a 2-tier audit across ALL products on a site (typically
200–5,000+ for mid-size to large Shopify stores), surfacing feed-risk gates,
traffic-weighted gaps, and a prioritized action queue.
Input Handling
- Site — e.g.
your-store.com(resolves to Shopify connector) - The pipeline assumes the Shopify connector is already authenticated for
the target shop. Verify with
get-shop-infobefore running.
Flags
--tier <variant-rich|technical-hardware|bundles-sets|consumables-bulk|seasonal-drops|specialty-premium>— restrict the run to one product-category tier (default: all tiers)--limit N(default: 30) — number of prioritized products to promote to full Tier-2 deep-scan viaproduct-analyze--min-impressions N(default: 0) — floor on 90d GSC impressions for a product to qualify for the opportunity ranking (use to exclude zero-traffic long-tail noise from Tier 2 selection; Tier 1 still scans everything regardless of this flag)--strict— pass-through toproduct-analyze; N/A items earn 0 instead of partial credit. Recommended default for pipeline runs.--no-gsc— skip GSC pull entirely (Tier 1 falls back to inventory/price signals only for prioritization; Tier 2 CTR sub-checks go N/A)--format html|md|json— default html for portfolio dashboard; md for human-readable; json for downstream consumption
Pipeline Phases
Phase 0 — Connector check (10 sec)
Verify Shopify connector is authenticated and matches the site:
get-shop-info → confirm domain matches argv[1]If mismatch, fail fast with "switch-shop required" instruction.
Phase 1 — Bulk metadata pull (2–8 min for ≤5,000 products)
Paginate search_products 50/page until hasNextPage = false. Cache to
output/[site]/products/_pipeline/shopify-products-all.json.
Fields collected per product:
id,handle,title,productType,vendor,tags,status,updatedAt,publishedAt,createdAtdescriptionHtml(length only, not full audit at this phase)seo.title,seo.descriptionvariants(count, price range,compareAtPrice,sku,barcode,inventoryQuantity,availableForSale)images(count, whether anyaltTextis null/empty)metafieldspresence flags (specs block, reviews app)
Phase 2 — Tier detection (in-memory, fast)
For each product, apply the tier rules from product-analyze SKILL.md
(hero / variant-parent / technical / consumable / new / standard), mapped
onto the kit's portfolio-level category taxonomy for reporting:
| Portfolio tier | Detection |
| variant-rich | ≥5 variants across color/size/style options |
| technical-hardware | productType/title implies spec-driven durables (electronics, tools, appliances, equipment) |
| bundles-sets | title/tags indicate multi-item SKU (kit, set, bundle, multipack) |
| consumables-bulk | low differentiation, repeat-purchase, low price, high inventoryQuantity turnover |
| seasonal-drops | tags/collection membership indicate a limited-window release |
| specialty-premium | high price point relative to store median + low variant count |
Phase 3 — Tier 1 quick-scan score (in-memory)
For each product, score against Shopify-metadata-only checks (no WebFetch needed):
| Check | Max | Detection |
| Description length adequacy | 6 | descriptionHtml word count ≥ 150 (else flag thin) |
| Image count adequacy | 5 | images.count ≥ 4 (else flag under-imaged) |
| Image alt-text coverage | 4 | 0 images with null/empty altText |
seo.title present + length | 4 | 50–60 chars, non-empty |
seo.description present + length | 3 | 120–160 chars, non-empty |
| Price sanity | 3 | price > 0; compareAtPrice (if set) > price |
| Identifiers present (GTIN/MPN/SKU) | 5 | at least sku; barcode present when vendor is a known manufacturer |
| Inventory / availability sane | 4 | availableForSale consistent with inventoryQuantity > 0 |
| Published + not orphaned | 3 | status = ACTIVE; not missing from every collection |
| Freshness | 3 | updatedAt within 180 days |
Total Tier 1 max: 40 pts. This is a cheap proxy, not the full 100-pt
product-analyze rubric — it exists purely to rank/triage before spending
WebFetch + GSC budget on Tier 2.
Phase 4 — GSC 90-day pull (bulk, per-URL)
Pull impressions/clicks/position/CTR for each product URL (batch query,
not one-by-one). Merge into the Phase 1 cache. Skip if --no-gsc.
Phase 5 — Site-wide pattern statistics
Compute and report on the full product set:
- % thin description (<150 words) by tier
- % under-imaged (<4 images) by tier
- % missing alt text
- % missing
seo.title/seo.description - % missing identifiers (
sku/barcode) by vendor - % zero-impression products (90d) — dead-weight candidates, separate from the optimization queue
- Tier distribution (variant-rich / technical-hardware / bundles-sets / consumables-bulk / seasonal-drops / specialty-premium counts)
- Stale-180d vs recently-updated split
Phase 6 — Merchant-Center feed-risk pre-scan 🛑 (highest priority signal)
Before ranking by opportunity, run a lightweight feed-risk check against the Phase 1+4 cache for every product (this does NOT require the full Tier-2 WebFetch pass — enough signal exists in Shopify connector fields alone to flag candidates for confirmation):
compareAtPriceset but ≤price(schema/price inconsistency risk)availableForSale = truebutinventoryQuantity = 0and no backorder/pre-order tag (availability mismatch risk)availableForSale = falsebut still linked from active collections/nav (stale disapproved-looking listing)- Missing
barcode(GTIN/MPN) AND vendor matches a known-manufacturer pattern (brand-name vendor, not a private-label placeholder) status = ACTIVEwith zero images (schemaimagerequirement failure)
Any product matching one or more of these earns a 🛑 FEED-RISK CANDIDATE
flag. These candidates are queued for Tier-2 confirmation first,
independent of their Tier 1 score or traffic — see Phase 7. A disapproved
product earns $0 regardless of how well-written its copy is, so this gate
overrides normal opportunity ranking.
Phase 7 — Opportunity ranking + Tier 2 selection
Rank all products (excluding --min-impressions floor failures, if set) by:
opportunity_score = normalized(gsc_impressions_90d) × normalized(gap_severity)
gap_severity = (40 − tier_1_score) / 40Then build the Tier-2 candidate list in this order:
- All
🛑 FEED-RISK CANDIDATEproducts from Phase 6 — top of the queue unconditionally, capped only by a sane ceiling (e.g. top 50 by revenue/traffic if the flagged set is very large) so the pipeline doesn't stall auditing a long tail of zero-traffic feed risks before anything else. - High traffic + weak page — top
opportunity_scoreproducts not already captured in (1). This is "high impressions, low Tier-1 score" — pages leaking the most existing demand. - Bottom-quartile Tier 1 score with meaningful traffic — fills out
the remainder of
--limitif (1)+(2) don't reach it.
Skip: status != ACTIVE, orphaned zero-collection zero-traffic products
(not real audit candidates — flag separately as "prune candidates" in the
report, don't spend Tier-2 budget on them).
Phase 8 — Tier 2 deep-scan dispatch
For each selected product, dispatch
product-analyze <url> --tier <auto-detected> --strict and collect the
full 100-pt scorecard, including the 🛑 Merchant-Center feed-risk GATE
detail from that skill's Step 3.
Delegate Tier-2 dispatches to sub-agents at the appropriate model tier rather than running them serially in the main session:
- Route each
product-analyzecall through a Sonnet-tier sub-agent (Agent(model: "sonnet")) — this is a "multi-step routine: apply rubric, pull 3 data sources, score, write file" task, not a judgment call. - Batch dispatches (e.g. 5–10 concurrent sub-agents) rather than one at a time, since each product's audit is independent of the others.
- Reserve the orchestrating session (this pipeline) for aggregation, ranking, and the final portfolio synthesis — not for running the per-product rubric itself.
If --no-gsc not set, mandate GSC pull for each (already cached from
Phase 4 — reuse, don't re-pull).
Phase 9 — Portfolio report assembly
Output:
output/[site]/products/_portfolio/product-audit-portfolio.md— main reportoutput/[site]/products/_portfolio/action-queue.md— ranked action list (feed-risk first, then opportunity-ranked)output/[site]/products/_portfolio/site-patterns.md— pattern statsoutput/[site]/products/_portfolio/prune-candidates.md— orphaned / zero-traffic products not worth auditingoutput/[site]/products/[handle]/product-quality-score.md— per Tier-2-selected product (from Phase 8 dispatch)
Phase pre-requisites (don't skip)
| Phase | Exit criterion |
| 0 | Connector authenticated + domain matches |
| 1 | All products in JSON cache, count ≥ 10 (else nothing to audit) |
| 2 | Every product has a portfolio tier assigned |
| 3 | Tier 1 scores in distribution |
| 4 | GSC merged (or explicitly skipped via --no-gsc) |
| 5 | Pattern stats computed |
| 6 | Feed-risk candidates flagged (list may be empty, but the pass must run) |
| 7 | Tier-2 candidate list built, feed-risk products first |
| 8 | Per-product full scorecards saved (for all selected candidates) |
| 9 | All report files written |
Cost estimate (rough, per mid-size site ≈ <N> products)
- Phase 1 (pagination): scales with product count / 50 per page
- Phase 2–3, 5–7: in-memory, no API calls
- Phase 4 (GSC bulk): 1 batched pull, not per-product
- Phase 6: in-memory pass over the Phase 1+4 cache, no extra API calls
- Phase 8 Tier 2 (default 30 products): ~30 × (1 Shopify + 1 WebFetch + 1
GSC) = ~90 API calls, dispatched via batched Sonnet sub-agents. Optional
narrower run with
--limitor--tier. - Phase 9: file writes only
Total: scales sub-linearly with catalog size thanks to the Tier 1/Tier 2
split — Tier 1 covers the whole catalog cheaply, Tier 2 spend stays capped
by --limit regardless of how large the store is.
What this skill does NOT do
- Does NOT optimize or rewrite product copy/schema, and does NOT push any
change to Shopify (audit-only) — that's
product-optimize - Does NOT replace
product-analyzefor single-page deep audits — that skill remains the canonical "I want to deep-dive THIS product" tool - Does NOT resolve feed-risk flags itself — it surfaces and prioritizes
them; confirmation + fix still runs through
product-analyze(Tier 2) andproduct-optimize
Composition
- Calls:
product-analyze(per prioritized product in Phase 8, via delegated sub-agents) - Feeds:
product-optimize— consumes the ranked queue + per-product reports as its work backlog - Reads from: Shopify connector + GSC connector
- Writes to:
output/[site]/products/_portfolio/+ per-product artifacts underoutput/[site]/products/[handle]/
Output convention
output/[site]/products/_portfolio/
├── shopify-products-all.json (Phase 1 cache)
├── tier-1-scores.json (Phase 3 output, all products scored)
├── site-patterns.md (Phase 5 stats)
├── feed-risk-candidates.md (Phase 6 flagged list)
├── action-queue.md (Phase 7 ranked list)
├── prune-candidates.md (Phase 7 excluded orphans)
└── product-audit-portfolio.md (Phase 9 main deliverable)
output/[site]/products/[handle]/
└── product-quality-score.md (Phase 8, Tier-2-selected only)Re-run / re-audit
Pass --resume to skip Phase 1 if the JSON cache is < 24h old. Useful when
iterating on report format, re-running Tier 2 with a different --limit,
or re-checking feed-risk candidates after a fix has been pushed via
product-optimize.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

