aeko-find-prompts-to-track

AEKO Find Prompts To Track

Helps the user discover research prompts worth tracking. Goal: after this skill runs, the user has 5-10 new tracked prompts relevant to their brand, without having to leave Claude and paste into the dashboard.

Marketer-facing output contract

Explain prompts as "questions your customers may ask AI." Keep ranking tables compact and show why each prompt matters: audience, buying stage, and competitor visibility. Never track prompts without explicit user selection.

Language: mirror the user's chat language for user-facing steps, summaries, questions, and confirmations. Keep slash commands, IDs, file paths, prompt metadata keys, and tool names in English/ASCII.

Input

• domain-id (optional) — UUID of the domain to scope suggestions to. If missing, offer aeko_list_domains pick-list.

Step 1 — Resolve domain + brand context

1. If $1 is set → use it.
2. Else → aeko_list_domains. If one domain: auto-pick. If multiple: show list and ask.
3. Call aeko_get_domain_info(domain_id) and aeko_get_brand_kit(domain_id). Use these to seed sensible default filters:
   • country = first entry in domain's selected_markets if available, else from target_country on the brand kit, else ask.
   • scope = domain's industry / vertical / scope field if set.

Step 2 — Ask user for filter intent

Prompt in the user's chat language. Use the matching KO/EN template below when applicable; for other languages, translate the EN template naturally while keeping platform/country/query-type tokens unchanged.

KO:

어떤 연구 프롬프트를 찾을까요? 비워두면 브랜드 기본값으로 검색할게요.

- 플랫폼: [claude / openai / google / perplexity / all]
- 국가: [KR / US / JP / ... / all]
- 페르소나 (예: new_mom, enthusiast, beginner)
- 키워드 (예: "이불 추천", "민감성 피부")
- 쿼리 유형 (informational / comparison / recommendation / transactional)

원하는 조합을 자유롭게 알려주세요.

EN:

What research prompts should I look for? Leave fields blank to use your brand defaults.

- Platform: [claude / openai / google / perplexity / all]
- Country: [KR / US / JP / ... / all]
- Persona (e.g. new_mom, enthusiast, beginner)
- Keyword (e.g. "bedding recommendation", "sensitive skin")
- Query type (informational / comparison / recommendation / transactional)

Describe the combination you want.

Parse the user's response into filter args. all / blank → omit that filter (no param to the tool).

Step 3 — Search

Call aeko_search_research_prompts with the parsed filters + page_size=25. At least one filter must be non-null (the tool rejects fully empty queries).

If zero results → widen: drop the most restrictive filter (typically persona_type or query_type) and retry once. Tell the user what was relaxed.

Step 4 — Score + rank candidates

For each returned prompt, assemble a relevance score using these signals:

• Keyword overlap between brand kit brand_keywords[] + must_include[] and the prompt's text or keywords[]. High overlap → high score.
• Persona match — prompt's persona vs brand kit target_audience.
• Latest-response signals (from the search payload's latest_response):
  • High mention_count across competitors → the prompt is a battleground worth contesting.
  • Presence of this brand in the mention breakdown → already getting mentioned; may not need tracking (user's call).
  • Absence of this brand with strong competitor mentions → prime tracking candidate.
• Funnel stage balance — prefer a mix across awareness / consideration / decision rather than all-same-stage.

Show top 10-15 sorted by score, in a compact table:

| # | Prompt | Platform | Country | Persona | Competitors mentioning |
|---|--------|----------|---------|---------|------------------------|
| 1 | ...    | Claude   | KR      | new_mom | 필리, 모노랩스          |

Include the prompt ID as a monospace UUID column so the user can reference specific rows.

Step 5 — Pick-list

Ask the user:

KO: "몇 번을 트래킹할까요? 쉼표로 여러 개 입력 가능. 전부 추적하려면 all 입력." EN: "Which ones should I start tracking? Comma-separated for multiple. all to track every row shown."

Parse response into a list of row indices → prompt payloads.

Step 6 — Pre-flight package limits

Call aeko_get_tracked_prompts to see the user's current tracked count. Compare against known package limits (derived from brand kit metadata.account_tier or, if unknown, just proceed and let the backend 403 — tell the user that's the fallback).

If the user's selection would exceed their package cap → warn, show how many they can track, ask them to narrow.

Step 7 — Track each selected prompt

For each selected row, call:

aeko_track_prompt(
    raw_prompt=row.raw_prompt,
    ai_platform=row.ai_platform,
    prompt_en=row.prompt_en,
    prompt_ko=row.prompt_ko,
    model=row.model,
    language=row.language,
    country=row.country,
    industry=row.industry,
    vertical=row.vertical,
    persona=row.persona,
    tags=row.tags,
)

Handle response codes:

• Success → note the new tracked prompt ID.
• 409 (already tracking) → skip, note "already tracked".
• 403 (package cap hit) → stop loop; tell user how many succeeded before cap.
• Reactivation message (backend returns 200 with status tracked when previously untracked) → note "reactivated".

Step 8 — Summary

Print:

✔ Tracking started for N prompts

  Newly tracked (N):
    - <prompt_id> · <ai_platform> · <country> · <first 80 chars>
    - ...
  Reactivated (M):
    - ...
  Skipped — already tracked (K):
    - ...

Next: AEKO will re-query these on cadence; check back in 1-3 days.
      For a deep-dive on one prompt's citation footprint:
      /aeko-prompt-deep-dive <prompt_id>

Error paths

• No domains connected → tell user to add one at the AEKO dashboard; stop.
• Search returned zero even after widening → suggest manual prompt creation via AEKO dashboard; stop.
• User picks nothing → exit cleanly, no writes.
• Backend 403 on track → surface the upgrade pitch from the error verbatim (backend includes the package-cap reason and pitch message).

What this skill never does

• Never tracks a prompt without explicit user selection.
• Never calls aeko_track_prompt with fields the search result didn't provide (no fabrication).
• Never deletes or untracks existing prompts — for that, use /aeko-prompt-deep-dive or the dashboard.