AppGrowing MCP Is Now Live

---
name: agg-single-product-report
description: This skill is applicable to: when the overseas mobile advertisement creative data of **single product/App** obtained based on AGG-MCP (searchAdCreatives, creativeAIAnalysisData, filterOptions, creativeResourceUrl) is required, when the Single Product Creative Strategy Diagnostic Report is produced. It focuses on a specific product (its own App or a single competitor), diagnoses its creative engine activity, selling-point focus, TOP creatives, creative life cycle and recession signal, and gives "amplification / iteration / elimination" creative asset disposal suggestions. Trigger scene: single product creative review meeting, own product creative health diagnosis, single competitor deep dismantling; or the user asked "help me systematically diagnose our XX game's creative play in the past month, which creatives should be added, changed or cut" / "deeply dismantle competitor YY's creative strategy in Japan".
agent_created: true
---

# AGG Single Product Creative Strategy Report

## Overview
AGG-MCP × Skill × Automation **single-scenario** analysis skill. It takes a normalized creative data set for a given product, converted into a structured **single-product creative strategy diagnostic report** (new scenario "F · Single-Product Creative Strategy Report"). Unlike "agg-category-report" (category level, cross-product), this skill only focuses on **one product** and does an in-depth physical examination of creative assets. This skill is decoupled from the MCP fetching layer: it consumes the `CreativeRecord[]` that is pulled and normalized by the MCP/orchestration layer in advance (recommended to use `searchDsl.campaign_name` to lock a single product), and does not call any tools by itself.

## When to call
- Single product creative review / own product creative health diagnosis (manual or Automation).
- Do a deep disassembly of one competitor (rather than a side-by-side comparison across competitors, which uses "agg-competitor-compare").
- Users ask questions such as "systematically diagnose the creative playing methods of our XX game in the past month, which should be added, changed, and cut" and "deeply dismantle competitor YY's creative strategy in Japan".

**Don't** use it to fetch data. If the `CreativeRecord` dataset is not available, run AGG-MCP first (filterOptions + creativeAIAnalysisData to lock dimensions → use `searchDsl.campaign_name` to lock a single product and pull searchAdCreatives, preferably covering multiple time slices), then return to this skill.

## Core Workflow
1. **Load contract**: Read `references/data_contract.md`, confirm required fields (`app_name`, `selling_points`, `format`, `ratio`, `language`, `impression`, `similar_cnt`, `advertising_days`, `max_dt`; ≥20 items in the product window).
2. **Issue analysis prompt**: Copy `references/analysis_prompt.md` and replace `<DATASET_SUMMARY>` / `<PARAMS>` (product name, region, time window), execute it.
3. **Apply deterministic rules** (see below, strictly as such):
   - **Creative engine activity**: total amount, number of new window items, `new_ratio`, and median of advertising_days.
   - **Selling-point focus**: TOP1 selling-point percentage; ≥40% judged "focused", <20% judged "scattered".
   - **TOP creatives**: impression / similar_cnt each take TOP5, mark "benchmark creative".
   - **Creative life cycle**: advertising_days distribution → distinguish evergreen (long cast) vs short-lived creatives; recession signal = few recent new items and old creatives still dominant.
   - **Disposal recommendations in three categories**: amplify (high performance and still worth casting), iterate (mediocre performance but form/selling points can be optimized), eliminate (low performance, long cast or declined).
4. **Delivery by envelope**: Fixed envelope (conclusion → evidence → recommendation → boundary). Recommendations given in "amplify / iterate / eliminate" three columns, each corresponding to specific creatives or selling points.
5. **Processing gaps**: When the sample is insufficient or a dimension is missing, only omit the corresponding subsection and mark at the boundary; no estimation allowed.

## Input/Output Contract (Abstract)
- **Input**: the `CreativeRecord[]` of a single product (time slices recommended to view evolution). Optionally with the same-category baseline for deviation comparison.
- **Output**: fixed-envelope Markdown insight block. Numbers must be calculated from the data set; if not available, write "insufficient data".

## Boundary Statement (key — must inform stakeholders)
This analysis uses **creative-side proxy metrics** (creative magnitude, similarity, exposure proxy, delivery days). It **is not** real delivery magnitude / consumption data (the latter requires `adVolumeTrends` / `regionOverview`). Each deliverable must be marked with: *"This analysis is based on creative-side proxy metrics, not true delivery magnitude."*

---
name: agg-competitor-compare
description: This skill is applicable to: when you need to obtain the overseas mobile advertising creative data of multiple competing apps through AGG-MCP (searchAdCreatives, creativeSellingPointTrends, filterOptions, creativeResourceUrl), when making a horizontal comparison of competitive creative delivery strategies. After normalizing the competitors' creative records, it produces structured cross-competitor comparisons — creative magnitude, new-launch rhythm (new ratio), media/format/language ratios, TOP selling points — and classifies each competitor as "adding party / changing party / keeping party" by hard rules, giving actionable follow-up recommendations for our side. Trigger scenario: competitor-monitoring weekly push (Automation) before the UA weekly meeting; or the user proposes "compare competitors XX and YY's buying creative methods this week" / "who is more radical, what selling points each promotes, and what are the channel and form differences".
agent_created: true
---

# AGG Competitor Creative Strategy Comparison

## Overview
Analysis skill for **single scenarios** within the AGG-MCP × Skill × Automation framework. It takes the normalized multi-competitor creative data set and converts it into a **competitor creative strategy comparison** report ready for the weekly meeting (scenario "A · Competitor launch strategy comparison push"). This skill is decoupled from the MCP fetching layer: it consumes the `CreativeRecord[]` pulled and normalized by the MCP/orchestration layer in advance; it does not call any tools by itself.

## When to Call
- Scheduled competitor-monitoring weekly push (Automation), used to synchronize intelligence before UA / buying weekly meetings.
- Users ask to compare competitors' creative buying methods, e.g. "Compare competitors XX and YY's buying methods in North America SLG: who is more radical? What selling points does each emphasize? What are the channel and form differences?"
- Any task that requires a side-by-side comparison of creative gestures for ≥2 competitors, rather than a single-brand perspective.

**Do not** use it to retrieve data. If there is no `CreativeRecord` data set, run AGG-MCP first (filterOptions → use `searchDsl.campaign_name` to pull searchAdCreatives for each competitor), then return to this skill.

## Core Workflow
1. **Load Contract**: Read `references/data_contract.md` and confirm the input matches the schema and this skill's required fields (`app_name`, `advertising_days`, `selling_points`, `media`, `format`, `language`, `similar_cnt`; at least 1 competitor ≥10 records).
2. **Issue the analysis prompt**: Copy `references/analysis_prompt.md`, replace `<DATASET_SUMMARY>` / `<PARAMS>` (competitor list, time window, target region), and execute it.
3. **Apply deterministic classification**: Compute `new_ratio = new-in-window count / total` per competitor; classify by hard rules — adding party `new_ratio ≥ 0.40` or highest absolute new count in the set; changing party `0.15 ≤ new_ratio < 0.40` with newer creatives; keeping party `new_ratio < 0.15`. Strictly follow this and do not invent categories.
4. **Delivery by envelope**: Wrap the output in the fixed envelope of `references/data_contract.md` (conclusion → evidence table → suggestion → boundary). Evidence table columns: "Competitor | Total | New ratio | Main media/format | TOP selling point | Classification". Give exactly 3 follow-up suggestions for our side.
5. **Processing gaps**: When a competitor's sample is below the lower limit, still output based on existing data and mark it in the boundary; no estimation allowed.

## Input/Output Contract (Summary)
- **Input**: `dict[app_name, CreativeRecord[]]` or flat `CreativeRecord[]` with `app_name`.
- **Output**: a fixed-envelope Markdown insight block. Numbers must be calculated from the data set; if unavailable, write "insufficient data".

## Boundary Statement (key — must be communicated to stakeholders)
This analysis uses creative-side proxy metrics (creative magnitude, new-launch rhythm, media/format/language ratios). It **is not** real delivery magnitude / consumption data (the latter requires `adVolumeTrends` / `regionOverview`, not in the current MVP). Each deliverable must be labeled: *"This analysis is based on creative-side proxy metrics and does not reflect actual ad-spend levels."*

---
name: agg-sellingpoint-trend
description: This skill is applicable to: when you need to obtain a category of overseas mobile ad creative data across two time slices based on AGG-MCP (searchAdCreatives, creativeSellingPointTrends, filterOptions, creativeResourceUrl), when summarizing "selling-point trend changes". It calculates the month-on-month ranking of selling points, marks emerging / declining selling points by fixed thresholds, and outputs a priority list for next month's creative topics. Trigger scenario: monthly creative topic-selection meeting; or the user asked "help me look at the creative selling-point trend of cross-border e-commerce last month: which selling points are rising and which are falling? What angle should I push next month".
agent_created: true
---

# AGG Category Selling-Point Trend Summary

## Overview
Analysis skill for **single scenarios** within the AGG-MCP × Skill × Automation framework. It converts two time-sliced category data sets into a structured **selling-point trend summary** (scenario "C · Category creative selling-point trend summary"). This skill is decoupled from the MCP fetch layer: it consumes the normalized `CreativeRecord[]` pulled in advance by the MCP/orchestration layer and sliced in time; it does not invoke any tools itself.

## When to Call
- Scheduled monthly trend briefing (Automation), used before creative topic selection.
- Users ask for selling-point heat trends, such as "last month's cross-border e-commerce selling-point trends: which are rising and which are falling? What should we push next month".
- Any task that needs to compare the changes of a category's selling-point pattern across two cycles.

**Do not** use it to retrieve data. If there is no data set, first run creativeSellingPointTrends (list all selling points) → searchAdCreatives for this category this month and last month respectively, then return to this skill.

## Core Workflow
1. **Load the contract**: Read `references/data_contract.md`, confirm required fields (`selling_points` and `impression` must cover **two** slices; ≥2 slices with ≥20 items each).
2. **Generate the analysis prompt**: Copy `references/analysis_prompt.md`, replace `<DATASET_SUMMARY>` / `<PARAMS>` (category, two slice definitions), and execute it.
3. **Apply deterministic labels**: Compute month-on-month creative-volume and ranking changes per selling point. Emerging selling point = growth ≥50% and last month's base <20; declining selling point = decline ≥30%. Combine emerging selling points with high-base steady-growth selling points to rank P0/P1/P2 priorities. Strictly follow the thresholds.
4. **Delivery by envelope**: Fixed envelope (conclusion → evidence table → recommendation → boundary). Evidence table columns: "Selling point | Last-month volume | This-month volume | Month-on-month | Ranking change | Label".
5. **Processing gaps**: When a slice is below the lower limit, still output based on existing data and mark at the boundary; missing slices cannot be estimated.

## Input/Output Contract (Summary)
- **Input**: category selling-point data grouped by two time slices (this month / last month).
- **Output**: a fixed-envelope Markdown insight block. Numbers must be calculated from the data set; if unavailable, write "insufficient data".

## Boundary Statement (key — must be communicated to stakeholders)
This analysis uses **creative-side proxy metrics** (creative volume, exposure proxy). It is not real delivery-level data (the latter requires `adVolumeTrends`). Each deliverable must be labeled: *"This analysis is based on creative-side proxy metrics and does not reflect actual ad-spend levels."*

---
name: agg-region-media
description: This skill is applicable to: when you need to obtain the "region × media" creative data of overseas mobile advertising based on AGG-MCP (searchAdCreatives, creativeSellingPointTrends, filterOptions, creativeResourceUrl), when analyzing the region × media creative ecosystem. It maps the region × media creative landscape, marks channel concentration, describes cross-channel creative style differences, and outputs media budget priorities and risks. Trigger scenario: quarterly media budget planning or before entering a new market; or the user proposes "we are ready to enter Latin America — analyze the channel focus and creative style differences of Brazil / Mexico same-category creatives on TikTok and Meta, and which way the budget should be tilted".
agent_created: true
---

# AGG Region × Media Creative Ecosystem Analysis

## Overview
Analysis skill for **single scenarios** within the AGG-MCP × Skill × Automation framework. It takes (region, media) cross-grouped creative datasets and turns them into a structured **region × media creative ecosystem analysis** (scenario "D · Region and media delivery strategy analysis"). This skill is decoupled from the MCP fetching layer: it consumes the normalized `CreativeRecord[]` pulled and tagged by the MCP/orchestration layer in advance; it does not invoke any tools itself.

## When to Call
- Scheduled quarterly media budget review, or one-time analysis before entering a new market (Automation / manual).
- Users ask for region/media creative differences and budget inclination, such as "entering Latin America, analyzing Brazil/Mexico creative differences on TikTok and Meta — how should the budget be tilted".
- Any task that requires a region × media creative map rather than a single-channel perspective.

**Do not** use it to retrieve data. If there is no data set, run filterOptions (area codes / media IDs) → cross searchAdCreatives by area × media, then return to this skill.

## Core Workflow
1. **Load the contract**: Read `references/data_contract.md`, confirm required fields (`region`, `media`, `format`, `selling_points`, `advertising_days`; ≥3 (region, media) units with ≥5 items each).
2. **Generate the analysis prompt**: Copy `references/analysis_prompt.md`, replace `<DATASET_SUMMARY>` / `<PARAMS>` (region set, media set, time window), and execute it.
3. **Apply deterministic rules**: For each (region, media) unit output creative volume / form share / TOP selling points; channel concentration — a media's share within a region ≥50% marked "highly concentrated", otherwise "scattered"; creative style differences — qualitative description based on each media's format/language/selling_points combination; budget priority + risk (e.g. a highly homogenized channel → recommend differentiation or cost reduction).
4. **Delivery by envelope**: Standard envelope (conclusion → evidence → recommendations → boundaries). Units with sample <5 marked "insufficient sample" at the boundary.
5. **Processing gaps**: For missing units only omit that unit and enumerate at the boundary; no imputation.

## Input/Output Contract (Summary)
- **Input**: `CreativeRecord[]` with `region` and `media` tags (can be crossed).
- **Output**: a fixed-envelope Markdown insight block. Numbers must be calculated from the data set; if unavailable, write "insufficient data".

## Boundary Statement (key — must be communicated to stakeholders)
This analysis uses **creative-side proxy metrics** (creative magnitude, form/media/language percentages). It is not real delivery-level / consumption or media cost data (the latter requires `adVolumeTrends` / `regionOverview` / media APIs). Each deliverable must be labeled: *"This analysis is based on creative-side proxy metrics and does not reflect actual ad-spend levels."*

---
name: agg-viral-alert
description: This skill is applicable to: when you need to obtain a new-creative set based on AGG-MCP (searchAdCreatives, creativeSellingPointTrends, filterOptions, creativeResourceUrl) and compare against a ≥7-day baseline, monitoring overseas mobile advertising "viral creative burst signals". It applies three hard threshold gates (similar-count above P95 / exposure surge / ≥3 independent Apps focusing on the same selling point within 48h), and only sends an alert when at least one gate is hit. Trigger scenario: daily / conditional viral-creative monitoring (Automation); or the user asked "help me keep an eye on Japan-market RPG these two days for suddenly emerging viral creatives or new selling-point plays — if any, remind me immediately".
agent_created: true
---

# AGG Viral Creative Monitoring & Early Warning

## Overview
Analysis skill for **single scenarios** within the AGG-MCP × Skill × Automation framework. It converts the "new-creative set + baseline" into a **viral-creative alert** decision (scenario "E · Latest viral creative monitoring alert"). This skill is decoupled from the MCP fetching layer: it consumes normalized `CreativeRecord[]` (new creatives + baseline) pulled in advance by the MCP/orchestration layer, and does not call any tools or download creatives itself.

## When to Call
- Scheduled daily / high-frequency conditional monitoring tasks (Automation). Silence shall be maintained when there is no signal.
- Users ask to keep an eye on sudden viral creatives or new selling-point plays, e.g. "keep an eye on whether Japan RPG has any sudden viral creatives in the past two days, remind me if there are".
- Any task where the correct output when there is no signal should be "no anomaly" rather than fabricated insight.

**Don't** use it to fetch data or download creatives. The MCP layer is responsible for fetching; after a hit the MCP layer resolves `creativeResourceUrl`. If there is no data set, run searchAdCreatives (isNew=1, order=max_dt_desc, limited category/region) + 7-day baseline first, then return to this skill.

## Core Workflow
1. **Load Contract**: Read `references/data_contract.md`, confirm required fields (`creative_id`, `similar_cnt`, `impression`, `selling_points`, `media`, `max_dt`, `app_name`, `is_new`; baseline ≥7 days history; current window ≥1 item).
2. **Generate the analysis prompt**: Copy `references/analysis_prompt.md`, replace `<DATASET_SUMMARY>` / `<PARAMS>` (category, region, monitoring window), and execute it.
3. **Calculate the baseline first**: Take the P95 of `similar_cnt` and the median of `impression` per media over the past 7 days. If the baseline is <7 days → directly return "no anomaly (baseline under 7 days, no alert)" and stop.
4. **Apply three hard gates (alert on any hit)**: ① single `similar_cnt` > baseline P95; ② `impression` ≥ 3× the median of the same window and media (or day-on-day ≥200% if a time series exists); ③ the same `selling_points` delivered by ≥3 independent `app_name`s within 48h. **No relaxation allowed**.
5. **Delivery**: When hit — envelope with Headline = signal type; evidence table columns "Creative ID + detail-page link | Selling point | Main casting channel | Signal type"; recommended actions: follow-cast / avoid-pit / disassemble. When no gate is hit — return **only** "no anomaly", never fabricate signals.

## Input/Output Contract (Summary)
- **Input**: "new-creative `CreativeRecord[]`" + "≥7-day baseline" with consistent shapes.
- **Output**: either "no anomaly" (or the baseline-insufficient variant), or a fixed-envelope alert Markdown block.

## Boundary Statement (key — must be communicated to stakeholders)
This analysis uses creative-side proxy metrics (similar count, exposure proxy, selling-point co-delivery). It **is not** actual campaign volume data (the latter requires `adVolumeTrends`). When an alert is triggered, it must be marked: *"This alert is based on creative-side proxy metrics, not real delivery levels."*