Detect Spam

Detect Spam Feature Requests

You are helping a WooCommerce Marketplace team member identify and silently flag spam feature requests for a specific product. Spam is anything that isn't a genuine product suggestion: promotional links, gibberish, off-topic content, scraped/SEO copy-paste, etc.

Spam is handled differently from other triage actions: no comment is posted. The skill only changes status to spam.

Read first:

• .claude/skills/shared/RULES.md — ID display, URL carry-through, HTML entity decoding, translation, plain-text comments, confidence labels.
• .claude/skills/shared/DISPLAY.md — pagination conventions.
• .claude/skills/shared/PHASE_LOOP.md — end-of-phase loop (standalone vs orchestrated mode).

Step 1 — Resolve the product

Follow .claude/skills/shared/RESOLVE_PRODUCT.md in full.

Step 2 — Fetch all open feature requests

Skip this step if invoked by the orchestrator. The orchestrator passes input_path — read FRs from there as JSONL.

Call wccom-feature-requests-list with product_id: <id>, status: "publish", and per_page: 100. Paginate until a page returns fewer than 100 items. Collect id, title, description, status, votes, date, url for every request.

If the first page returns 0 results, stop and report: "No open feature requests found for this product."

Step 3 — DETECTION STARTS HERE — Identify spam

(Orchestrator subagent: begin reading from this step.)

A request is likely spam if any of these patterns apply:

• Promotional / link-stuffed — body is mostly URLs, especially to unrelated products, services, or shady domains; affiliate-style copy.
• Off-topic — about a different product, an unrelated industry, cryptocurrency, weight loss, gambling, adult content, "buy followers", prescription drugs, etc.
• Gibberish / random text — keyword soup, broken grammar that looks machine-generated or pasted out of context, lorem-ipsum-like filler.
• Scraped / SEO copy-paste — boilerplate marketing prose with no specific feature ask.
• Exact or near-identical duplicates — same body text across multiple requests; strong signal of coordinated spam.
• Account / contact info dump — phone numbers, emails, login credentials, "call us at…" style content.
• Trojan horse pivot — opens with a generic relatable statement then pivots to promote an unrelated product or link.
• Link syntax probing — lists of URL format variants (BBCode, markdown, wiki syntax) with no actual feature request content.
• Non-Latin script with no product relevance — content in a script unrelated to the product's audience that contains no genuine feature ask, especially paired with promotional links.

A request is NOT spam (leave it alone) if it:

• Describes any genuine product feature, no matter how rough or short.
• Is a support question (use /detect-support-requests).
• Is in a non-English language but on-topic — translate, don't flag.

Be conservative. False positives mean a real merchant gets silently disappeared. Only flag where the spam framing is unambiguous.

Assign confidence per RULES.md (High / Low).

Flagged record schema (for orchestrator output JSONL)

{ "id": <int>, "title": "<str>", "url": "<str>",
  "reason": "<one sentence>", "confidence": "High" | "Low",
  "excerpt": "<first 1–3 sentences of description, trimmed>" }

Step 4 — Present the report

Follow DISPLAY.md pagination. Decode HTML entities. Translate non-English excerpts per RULES.md.

For each flagged request:

## ID [id] — "[title]"
Confidence: High / Low
Reason: [one sentence — what makes this look like spam]

[votes] vote(s) · opened [date]
[url]

Excerpt: "[1–2 sentences from the description showing the spam framing —
verbatim, trimmed if long]"

End with a summary line: X open requests scanned · Y likely spam (Z high confidence, W low confidence).

If no requests are flagged, report that and stop.

Step 5 — Confirmation menu

Which requests should I mark as spam?

  [1] "[short title]" (confidence)
  [2] "[short title]" (confidence)
  ...
  [A] All of the above
  [H] High confidence only      ← omit if no High items (see RULES.md)
  [N] None / skip

Reply with numbers (e.g. "1 3"), A for all, H for high confidence only, or N to skip.

⚠️  Marking as spam is silent — no comment will be posted on the request.

Wait for the user's reply before taking any action.

Step 6 — Mark approved requests as spam

For each approved request:

1. Call wccom-feature-requests-update-status with id set to the FR's ID and status: "spam".
2. Fallback — if the API returns a permission error (403 / "not allowed"), retry with status: "closed". Record the fallback so the summary distinguishes it.
3. Write failure handling — if both attempts fail, do not record this FR as actioned. Report the failure in the summary and continue to the next FR.

Do not post any comment. Spam handling is silent by design.

Step 7 — Confirmation summary

Done.

🚫 Marked as spam: ID [id] — "[title]" ([votes] vote(s) · opened [date])
   [url]

❌ Closed (spam fallback — no spam permission): ID [id] — "[title]"
   [url]

⚠️  Failed: ID [id] — "[title]" — [error]
   [url]

⏭️  Left open (not confirmed): ID [id] — "[title]"

Omit any section with no entries. End with: N marked as spam · M closed as fallback · F failed · P left open.

Then follow PHASE_LOOP.md.