VGC Team Guide Generator

Claude supplements with strategic reasoning that the data doesn't directly provide: how team members interact, which leads work against which archetypes, tempo and pivoting logic, win condition identification. These require judgment, not lookup. The boundary is: if it's a fact (stat number, type effectiveness, ability description), look it up. If it's a strategic assessment (how to play the team, what to bring against TR), reason from the data.

Input Types

This skill accepts three input formats:

1. Pokepaste URL — e.g., https://pokepast.es/abc123. Fetch and parse the Showdown-format team.
2. Team JSON file — a file under knowledge-base/pokepastes/{SEASON}/ (e.g., knowledge-base/pokepastes/M-A/0516edwards-...-team.json). Read directly — this is the richest input source because it contains structured data including EVs, moves, abilities, items, and replica codes.
3. Chinese team HTML — a file under knowledge-base/pokepastes/{SEASON}-zh/ (e.g., knowledge-base/pokepastes/M-A-zh/0516edwards-...-team.html). Find the corresponding JSON file in the non--zh sibling directory (same filename, .json extension) and use that as input. The HTML is a rendered view — the JSON has the raw data the analysis needs.

When the input is a local file, the season ID is derived from the directory name (e.g., M-A). When the input is a pokepaste URL, read knowledge-base/regulations/index.json to determine the current season.

Workflow

Step 1: Resolve input and extract team data

Based on input type:

Pokepaste URL: Fetch the page. If the URL returns a non-200 status or the content doesn't contain Showdown-format team data, report the error and ask the user to verify the URL. Parse each team member's species, item, ability, nature, EVs, moves from Showdown format. Build comma-separated species and moves lists.

Team JSON: Read the file. Verify it contains a team array with at least one entry. If the team array is missing or empty, the file may be from a failed fetch — report this and suggest re-running fetch-pokepastes. Extract from team[] array: each member's species, item, ability, nature, evs, moves. Also capture team_id, player_name, description, season, pokepaste_url, and replica for the output header.

Chinese HTML: Derive the season from the directory path (M-A-zh → M-A). Construct the JSON path by replacing -zh in the directory and changing extension to .json. If the JSON file doesn't exist at that path, search the M-A/ directory for a file matching the same slug prefix. If still not found, report the error and suggest the user provide the team JSON or pokepaste URL directly. Read the JSON file and proceed as above.

Step 2: Determine season and run query

Season is mandatory — every guide must include regulation compliance analysis because an illegal team is useless in competitive play.

• If season is known from the input file (local JSON/HTML), use it directly.
• If input is a pokepaste URL, read knowledge-base/regulations/index.json and use the latest season.
• If no regulation data exists for the season, warn the user and proceed without regulation checks.

python3 .claude/skills/get-vgc-team/scripts/query_kb.py \
  --species "Species1,Species2,..." \
  --moves "Move1,Move2,..." \
  --season {SEASON_ID} \
  --kb knowledge-base

Pass all 6 species names (comma-separated) to --species, and all 24 move names to --moves. Read the full output — it contains the factual foundation for every analysis dimension.

The output includes: pokemon stats, move details (Chinese names, types, power, descriptions), ability descriptions, type chart analysis, field effects, meta overlap, relevant articles, and regulation data. If --season is provided and regulation data exists, the output includes:

• available_showdown_ids: legal Pokemon in this format
• mega_evolution: mega list and count
• regulation_warnings: team members NOT available in this format

Step 3: Generate the guide

Using the JSON context from Step 2, produce the guide following the Analysis Framework and Output Template below.

Step 4: Polish the summary with humanizer-zh

The 队伍总结 section will be shared with other players, so it needs to sound like a real person, not AI output. After writing the full guide:

1. Extract just the 队伍总结 section text (the markdown content under the ## 队伍总结 heading)
2. Call the Skill tool with skill: "humanizer-zh", passing the extracted text as the args parameter
3. Replace the 队伍总结 section content with the polished text returned by humanizer-zh
4. Keep the ## 队伍总结 heading unchanged — only the body text gets replaced

The goal: natural, conversational Chinese that conveys genuine insight — not stiff, not preachy, not overly casual. Think of how an experienced player would describe the team to a friend: direct, specific, with personality but grounded in facts.

Step 5: Save output and update index

Save the guide to knowledge-base/pokepastes/{SEASON}-guides/{team-slug}.md.

The output directory follows the existing naming convention — M-A for raw data, M-A-zh for Chinese pages, M-A-guides for strategy guides. This keeps all data for a season discoverable in one place.

Use the same slug as the corresponding team JSON/HTML file (e.g., 0516edwards-charizard-y-whimsicott-team.md). If the input is a pokepaste URL without a local file, generate a slug from the team description: lowercase, replace spaces/special chars with hyphens, max 60 chars.

After saving, update or create knowledge-base/pokepastes/{SEASON}-guides/index.json:

[
  {
    "slug": "0516edwards-charizard-y-whimsicott-team",
    "team_id": "PC60",
    "player_name": "0516edward",
    "description": "Charizard Y + Whimsicott Sun Team",
    "species": ["Charizard-Mega-Y", "Whimsicott", "..."],
    "style": "sun hyper offense",
    "season": "M-A",
    "generated_at": "2026-04-16T12:00:00"
  }
]

Append new entries to the array. If an entry with the same slug already exists, replace it. Sort by team_id. This index lets the site generate a season guide listing page and link from Chinese pages to their corresponding guides.

Analysis Framework

Nine dimensions from VGC competitive strategy. Each requires specific data lookups — reference the query_kb.py output to ground every claim. Season regulation data must be integrated throughout, not just in the overview.

A. Team Overview

Identify the team's identity in one paragraph:

• What style is this team? (hyper offense / balance / semi-room / weather / tailwind-based / trick room / ...)
• What's the team's "engine" — the combination that makes it work?
• How dependent is the team on a single win condition? (single-mode vs multi-mode)
• Regulation compliance: Check regulation.regulation_warnings. If any team member is not available in the current season format, flag this prominently — the team may need adjustment for legal play.

Data: pokemon stats, abilities, field_effects, speed distributions, regulation data.

B. Speed Tiers

Speed is the most important resource in VGC:

• Each member's base Speed and meta positioning
• Speed control methods (Tailwind, Trick Room, Icy Wind/Electroweb, Thunder Wave, weather abilities)
• Turn count of speed control and what happens when it expires
• Can the team operate without speed control?

Present a speed tier table sorted by base Speed.

C. Typing and Defensive Synergy

• Type core (e.g., Fire/Water/Grass, Fairy/Steel/Dragon)
• Shared weaknesses — types hitting 2+ members super effectively
• Defensive holes — types no member resists
• Cross-coverage through switches

Data: type_chart from query output (shared_weaknesses, resistance_coverage).

D. Core Combinations and Leads

• Core duos: Which pair is strongest and why? (redirection + setup, Fake Out + attacker, weather setter + abuser)
• Lead combinations: Turn-1 leads for different matchups
• Ability combos: Direct enablers (Intimidate + Defiant, weather + speed boost)
• Item-move combos: Weakness Policy + KO partner, White Herb + Contrary/Intimidate

Data: ability_details, move_details, field_effects.

E. Pivoting and Tempo

• Pace: hyper offense (~4 turns), balance (~8), grindy (12+)
• Switching logic: Who absorbs hits for whom
• Resource timing: Mega evolution turn, Tailwind/Trick Room turn, Protect stalling
• Fall-off point: Late-game plan if the team doesn't close early

F. Win Conditions

• Primary and backup win paths
• Key survivors that must stay alive
• Priority kill targets on the opponent's side

G. Team Selection Framework

For each major archetype, recommend 4 to bring:

Data: meta_overlap from query output.

H. Meta Matchup Analysis

• Champion teams sharing species with this team (same core = well-tested archetype)
• Favorable and unfavorable type matchups against meta
• 2-3 most threatening meta Pokemon/cores

Data: meta_overlap (species overlap counts, team descriptions).

I. Trick Room Preparedness

• Can this team prevent Trick Room? (Taunt, Fake Out, KO potential)
• If Trick Room goes up, how bad is it?
• Stalling strategies (Protect, defensive switching)
• Own Trick Room mode as backup?

Output Template

All text in simplified Chinese. Use markdown formatting. The template below shows the final output structure — the 队伍总结 section will contain polished text after the humanizer-zh step.

# {队伍名称} — VGC 队伍指南

> 输入: {source description — pokepaste URL, or team JSON path, or Chinese HTML path}
> 赛季: {season_id + date range, or "未知"}
> 选手: {player_name if available, or "未知"}
> 合规: {全部合规 / 以下成员不在当前赛制可用列表中: ...}

## 队伍概览

{一段话总结队伍风格和核心引擎}

| 位置 | 宝可梦 | 属性 | 特性 | 道具 | 定位 |
|------|--------|------|------|------|------|
| 1    | ...    | ...  | ...  | ...  | ...  |

## 速度线分析

| 宝可梦 | 基础速度 | 速度定位 | 控速手段 |
|--------|----------|----------|----------|
| ...    | ...      | ...      | ...      |

{速度控制分析段落}

## 属性与联防

{属性核心、共同弱点、联防覆盖分析}

## 核心搭配

{核心配对、首发组合、特性/道具联动}

## 轮转与节奏

{节奏定位、换人逻辑、资源时机}

## 胜利条件

{主/备胜利路径、关键存活、优先击杀}

## 选出框架

| 对阵类型 | 选出 | 排除 | 首发 | 理由 |
|----------|------|------|------|------|

## 环境应对

{对照冠军队伍的优劣势、环境热门对策}

## 戏法空间对策

{防空间、空间下存活、反空间手段}

## 队伍总结

{2-3 段话简要概括队伍核心玩法、使用要点和注意事项}

Writing Guidelines

• Write for intermediate VGC players — assume they know what EVs, natures, and STAB are
• Every claim must be traceable to query_kb.py output data. If data doesn't support a claim, don't make it
• Be specific, not generic. "Garchomp outspeeds most non-Mega meta Pokemon at base 102" is good. "Garchomp is fast" is bad
• Use Chinese for all narrative text. Format names as: 中文名 (English Name) on first mention, then just 中文名
• When uncertain, say so explicitly rather than guessing
• Don't pad with obvious observations. "Fire resists Grass" isn't worth stating unless it matters for this team's synergy

Summary-specific guidelines

The 队伍总结 section has different voice requirements from the rest of the guide:

• The analytical sections (A-I) should be precise and data-grounded — clinical is fine
• The summary should read like an experienced player talking to a peer: direct, specific, with personality but grounded in facts
• Avoid stiff academic tone and avoid overly casual slang
• Before humanizer-zh: draft naturally. After humanizer-zh: verify the polished text still accurately reflects the team analysis

Example Interactions

Example 1: User provides a pokepaste URL

User: "帮我看看这个队伍 https://pokepast.es/abc123"

Agent fetches the pokepaste URL → parses Showdown format → reads regulations/index.json → finds current season M-A → runs query_kb.py with extracted species and moves, --season M-A → generates full guide with all 9 analysis dimensions → extracts 队伍总结 text → invokes /humanizer-zh → replaces summary → saves to knowledge-base/pokepastes/M-A-guides/ → updates index.json.

Example 2: User provides a local team JSON

User: "分析一下 knowledge-base/pokepastes/M-A/0516edwards-charizard-y-whimsicott-team.json"

Agent reads the JSON → extracts team data (species, moves, etc.) and metadata (season=M-A, player_name, team_id) → runs query_kb.py with --season M-A → generates guide → the input header shows the JSON path and player info → polishes summary → saves alongside existing guides in M-A-guides/.