pokemon-champions-meta

Offline-first Pokémon Champions metagame queries for current/historical seasons and single/double formats, including usage, common configurations, partners, trends and competitive Top Meta. Use for "环

SKILL.md

GitHub

Pokémon Champions Meta

Query the bundled Pokémon Champions metagame snapshot for rankings, common configurations, partners, format comparisons, and factual changes between snapshots. This skill answers "what is popular and how is it used?"; canonical battle facts remain the responsibility of $pokemon-champions-dex.

Workflow

  1. Use the current season/rule unless the user explicitly requests a historical context.
  2. Use ranking for the in-game environment overview and detail for one Pokémon's panels. Use topmeta for the competitive real-team population and its differences from in-game usage. High-level wording always uses --scope high: 高端、高分段、高水平、天梯顶端、上位、比赛、大赛; high-end, high ladder, top ladder, high-rated, top cut, tournament; 上位帯、上位勢、高レート、大会、入賞. Never substitute the whole ladder for this population.
  3. Use who-uses to find which Pokemon run a move/item/ability/nature, and compare for single-vs-double ranks and per-Pokemon panel shares. search handles more general filters.
  4. Use ko for the knock-out axis — whom a Pokémon knocks out and who knocks it out.
  5. Use trend for one Pokemon's stored rank/panel history, movers for this season's top-30 risers/fallers, and report for changes since the latest snapshot. Use export-excel only when the user requests workbooks. For team-level format structure (speed control, signals, Mega slots and common pairs), use the sibling team skill's landscape --game-format <single|double> --format json; it reads real joint teams, not in-game usage marginals. Add --scope high for high-level structure; its sample counts qualifying published records, not Top Meta's player-deduplicated units.
  6. Keep usage marginals factual: common moves/items/partners are not a guaranteed joint set.
  7. When asked to update only meta, run python dev/update/update.py meta from the repository root. The default flow collects full single and double usage, KO objects and both KO move-share panels together. Do not ask for separate KO or move collection. Use refresh when real teams also need updating; developer source and backup operations are documented in dev/update/README.md.

Commands

Query cached rankings:

# Uses the current season/rule by default.
python scripts/meta_query.py ranking --format single --limit 20
python scripts/meta_query.py ranking --format double --limit 20
python scripts/meta_query.py ranking --format double --season M-4 --rule M-B --limit 20

Query Pokémon details:

python scripts/meta_query.py detail --format single --pokemon 雷丘
python scripts/meta_query.py detail --format double --pokemon garchomp
python scripts/meta_query.py inference-prior --format double --pokemon "Mega Garchomp Z"

Reverse-search panel entries or compare formats:

python scripts/meta_query.py search --format double --panel moves --name 地震
python scripts/meta_query.py search --format both --panel moves --type ground --category physical --min-usage 20
python scripts/meta_query.py search --format single --panel moves --where '{"and":[{"type":"Fire"},{"or":[{"category":"Special"},{"usage":">=20"}]}]}'
python scripts/meta_query.py compare --pokemon 雷丘
python scripts/meta_query.py who-uses --kind item --entry 讲究围巾 --format double
python scripts/meta_query.py trend --pokemon 烈咬陆鲨 --format single
python scripts/meta_query.py movers --format double
python scripts/meta_query.py report --format both

Query the KO axis (a separate data family with its own upstream snapshot):

python scripts/meta_query.py ko --format double --pokemon 大狃拉
python scripts/meta_query.py ko --format single --pokemon Salamence --panel koed_by
python scripts/meta_query.py ko --format single            # coverage only: what this snapshot has

ko_targets (alias beats) is whom the Pokémon knocks out; koed_by (alias counters) is who knocks it out. Both are an ordering only — the source publishes no percentage, so none is shown. Both are species-level: opponents are keyed by national dex number with no form, and a species that appears twice in one list is flagged rather than de-duplicated. ko_moves / koed_by_moves are a reserved tier: while coverage.move_share is absent they are null, meaning NOT COLLECTED — not "no KO moves". Never present a KO list as a matchup win rate; it counts knock-outs, not games.

Names resolve through the sibling dex in Chinese, English, or Japanese. Fuzzy corrections are always disclosed: structured detail/compare/ko results carry resolution metadata, while search reports a correction on stderr without contaminating JSON stdout. The public dex resolver accepts compositional Mega names such as Mega耿鬼 and メガゲンガー; explicit forms never fall back to a base row when the snapshot lacks that form.

Competitive Top Meta

python scripts/meta_query.py topmeta --format double --scope high --pokemon 烈咬陆鲨
python scripts/meta_query.py topmeta --format single --scope rule --limit 10
python scripts/meta_query.py topmeta --format double --scope season --pokemon Garchomp --view full

The default is compact canonical JSON (--view summary, display limit 6). --view full returns untruncated data. Always name the population, scope, sample and comparison window. high states its separate tournament/ladder criteria and composition; it is a filter, not a strength score. Read queries[] and projection.omitted before making claims. An unrepresented or ambiguous query stays visible. Published counts aggregate Megas into species: a Mega-only question reports the granularity gap and a possible base-name reference, without presenting that aggregate as Mega data.

This command has a read-only dependency on the sibling $pokemon-champions-team published data/top_meta/ tables and its pure comparison/projection modules. It does not load the team orchestrator, read the real-team library, rebuild data or call a network service. Missing team installation is explicitly unavailable; other meta commands continue to work without it. The data remains the same library evidence used by repset/frame, so agreement is not corroboration.

Use JSON for downstream processing:

python scripts/meta_query.py detail --format double --pokemon garchomp --output json

Contract And Output

Treat the executable schema as the authority for commands, flags, fields, boolean query grammar, and errors:

python scripts/meta_query.py schema

Canonical JSON is language-independent. Human-readable output language precedence is --lang, then POKEMON_CHAMPIONS_LANG, then en. Read references/api.md for detailed command usage and references/schema.md only when the bundled cache layout matters.

Excel Export

export-excel always writes three standalone single-language workbooks (zh, ja, en) to the chosen output directory, independently of --lang:

python scripts/meta_query.py export-excel --season M-4 --rule M-B

This is the only command requiring openpyxl; install requirements.txt when export is needed. All query commands use the Python standard library.

Boundaries

  • The shipped snapshot is read-only and queried offline; its date and season/rule stamp define freshness.
  • Do not infer legality or learnsets from usage data.
  • Do not combine move, item, ability, nature, partner, or spread marginals into an asserted joint set.
  • A change report records factual differences between snapshots, not an interpretation of the metagame.
  • Usage and KO normally share a complete snapshot; backups or retained KO can have different clocks. Quote each one's own updated_at and never treat a KO ordering as a win rate, a counter verdict, or a damage result. For zero-record directions retained from history, quote each panel's donor clock; feedback evidence and upstream reports use English.