pokemon-champions-team

Pokémon Champions single/double team building, legality validation, and diagnostics. Use for producing, completing, changing, reviewing, or evaluating a team; do not trigger from isolated words like c

SKILL.md

GitHub

Pokémon Champions Team

Build, validate, diagnose, tune, and review Pokémon Champions teams. This skill is an orchestrator: its scripts emit deterministic facts, legality verdicts, and auditable evidence; the model makes and explains strategic trade-offs. Never invent a composite team-strength score or an objective "best" team.

Use the sibling skills as factual authorities:

  • $pokemon-champions-dex: names, roster, forms, types, stats, abilities, learnsets, and items.
  • $pokemon-champions-meta: current/historical usage, common configurations, partners and competitive Top Meta (meta_query.py topmeta; high-level wording uses --scope high).
  • $ncp-damage-calculator: exact damage, KO/survival, and speed calculations.

Task routing

Read references/assisted_workflow.md for every whole-team request. It owns the UEP gates and evidence requirements; the table below selects the applicable path.

Task Entry and evidence
New build or complete a partial team context-audit → intake --next (until done; re-audit changed context) → ranking and joint evidence → frame → model candidates / slate-init → slate-evaluate → checkpoint → draft-init / answer-audit
Diagnose an existing team context-audit → review; one completeness-gated battery supplies diagnosis, selection and matchup facts. review-claims verifies scoped claims. No new-build intake/frame for a read-only diagnosis.
Rework, tune, or evaluate candidates Audit context, diagnose the original when present, ground changes in relevant joint evidence, and run tune for explicit adjustment benchmarks. Slate the resulting proposals and audit the answer. A new-build frame is required only when building/completing a team.
Single damage/speed fact Use the sibling calculator; no whole-team build flow.
Type effectiveness or current rule fact type-matchup and rules-facts return fixed type-chart and reviewed rule facts. These factual queries do not enter the whole-team UEP flow.
Speed against the metagame speed-ladder <species> --game-format single|double; optional stated set and conditions. This factual query does not enter the whole-team UEP flow.
Who answers a Pokémon counters <species> --game-format single|double --top-n 30; retained build facts, no whole-team UEP. Keep conditional C0 blockers and missing evidence in the answer.
Metagame OHKO/2HKO table ko-scan <species> --game-format single|double; both directions by default, explicit reference build and qualification groups. Use the calculator for a stated set or one exact pair.
Stated scenario on common builds scenario scenario.json; both combatants use exact-form retained builds with explicit set and field overrides. Read its source and conditions before citing a KO.
SP threshold on common builds threshold threshold.json; minimum legal SP for a stated KO, survival or speed goal, with nature/item lanes, closest feasible result and optional secondary SP maximization.
Retrieve observed evidence or Top Meta repset, observed, search / show, landscape, topmeta; keep samples and in-game usage distinct.

Normalize user names through the dex and construct the JSON inputs yourself. Read references/schema.md §1 for authored-set completeness, Mega representation and canonical SP keys. Use actual battle-form names in tune benchmarks, replacement selectors and frame deviations: Mega Charizard X selects a Charizard holding Charizardite X; Charizard selects its ordinary form. Nested names resolve through the public dex; fuzzy corrections are disclosed and unknown names must be fixed before running the operation. Original inputs remain unchanged for receipt binding. Save full gate outputs unchanged: receipts bind the context, frame and candidates. Never present an invalid, eliminated or unaudited proposal. checkpoint normally lets candidate comparisons proceed; it asks for an intermediate choice when requested (checkpoint_required:true) or when modal Mega-registration coverage needs a repair or an explicit posture. Missing benchmarks are disclosed, not asked for automatically.

slate-init transports explicit candidates, selected frame primary sets, bindings and resolved intake IDs into a directly usable slate. draft-init fills derivable onboarding facts and offers receipt-bound claim_candidates; neither helper chooses the team or writes its strategic rationale. The complete input/round-trip example is in references/schema.md §8.

Use build-context.goals:[{kind:answer|include,species:...}] for named opponent/member targets. Slate includes answer targets outside its top-K and returns receipt-bound comparison facts: CHECK floors, holes, candidate/core overlaps and library overlap, without ranking candidates. Read coverage and incomplete-calculation qualifications. If cores overlap over half or cannot be compared, fill draft.exploration.reason; library adoption also discloses whether other structures were considered. Team matchup claims use team_claims with complete saved report rows. Use localize team.json for official trilingual member/set names. Optional draft.text_check:{lang:zh|ja|en,text:...} checks the final text's known names, member presence, percentage claims and writing script; it does not certify prose semantics. See schema §8.

For alternatives, use frame's --draw N --seed INT --seen-teams shown.json or observed --order diversity. fill accepts need.answer to retrieve complete observed builds answering named opponents, with static CHECK conditions; the off-meta view also includes library-backed long-tail species. See references/schema.md §8 for evidence and identity limits.

The explicit keyword 建队流程 and its aliases enter the complete guided path unconditionally.

Contract Discovery

Do not guess payload shapes. The CLI exposes its commands and structured contracts:

python scripts/team.py schema
python scripts/team.py vocab

schema describes command inputs, team-json, context specs, session specs, and answer drafts. vocab describes enumerated intent fields and which fields are enforced by scripts versus interpreted by the orchestration layer. Read references/schema.md for durable domain and cache contracts.

Core Commands

An explicitly supplied ten-character public team code is a team input. Resolve it once with python scripts/team.py code <code> --format json, then use its team in the existing workflow. Never guess or enumerate codes, fetch while typing, or retry a refusal automatically. --cache-only makes no network request. A query does not establish format, season or rule: preserve nulls until the caller supplies the context. Query failure must not become a fabricated or species-only team. Network setup, conservative shared limits and error contracts: references/team-codes.md. The resolved team still goes through the normal validation/build/diagnostic gates.

Parse and validate:

python scripts/team.py parse team.txt --game-format double --format json
python scripts/team.py validate team.json --context ctx.json --format json
python scripts/team.py suggest-fixes team.json --context ctx.json --out suggestions.json
python scripts/team.py diagnose team.json --aspect all --format json

suggest-fixes returns explicit ability/item choices and every side of an Item Clause conflict, without editing the team. Choose and disclose changes on a copy, then rerun validation and the affected UEP gates. Item options are a verified convenience shortlist, not a suitability ranking; occupied/avoided options are marked unavailable. See references/schema.md §16.

Selection, matchup, and tuning:

python scripts/team.py review team.json --context ctx.json --out review-full.json
python scripts/team.py try team.json --changes changes.json --out trial-full.json
python scripts/team.py diagnose team.json --focus targets.json --format json --out focused-review.json
python scripts/team.py review-claims claims.json --fail-on-verdict
python scripts/team.py select team.json --context ctx.json --format json
python scripts/team.py matchup team.json --top-k 20 --context ctx.json --format json
python scripts/team.py matchup actual-sets.json --top-k 60 --matchup-view summary --format json --out matchup-full.json
python scripts/team.py tune team.json --context ctx.json --format json
python scripts/team.py fill team.json --context need.json --slot 1 --format json --out replacements.json
python scripts/team.py review team.json --opponents opponents.json --conditions field.json --out scoped-review.json
python scripts/team.py select team.json --opponents opponents.json --conditions field.json --format json
python scripts/team.py tune cell-spec.json --from-cell --out cell-tuning.json
python scripts/team.py type-matchup --attack-type 地面 --pokemon Rotom-Wash --format json
python scripts/team.py type-matchup --types Steel Fairy --ability Levitate --format json
python scripts/team.py rules-facts --format json
python scripts/team.py speed-ladder "Mega Gengar" --game-format double --limit 20 --format json
python scripts/team.py speed-ladder Garchomp --game-format double --set speed-set.json --conditions speed-field.json --opponent-ranges --format json --out speeds.json
python scripts/team.py counters Garchomp --game-format single --top-n 30 --view summary --format json --out counters-full.json
python scripts/team.py ko-scan "Mega Garchomp Z" --game-format double --top-n 30 --view summary --format json --out ko-full.json
python scripts/team.py scenario scenario.json --out scenario-result.json
python scripts/team.py threshold threshold.json --out threshold-result.json

review and try default to compact stdout while --out and session references keep the full receipt-bound result. Missing moves, ability, nature, SP or trusted source evidence returns needs_input without team CHECK/selection claims. try accepts explicit {slot,set} changes; it edits a copy, replaces whole fields, and clears the old set when species changes. Read the structural differences and both qualified floors; the tool does not choose a winner. --focus maps {members:[name|slot],opponents:[name]} to slots and evidence coordinates, adding named opponents beyond top-K. For claims, copy a full report's evidence_index reference and complete fact into review-claims; it recomputes team floors, opponent variants and member cells without mixing their scopes. A successful check does not replace proposal UEP gates. See schema §17.

fill --slot N releases that zero-based member's species and item before retrieval. It returns complete observed replacement configurations, excludes retained/avoided items, and attaches changes ready for try, with structural differences and actual set Speed. Empty results disclose omitted evidence; do not invent missing SP or treat structural changes as matchup improvements.

--opponents supplies 1–6 species-only entries (observed variants) or complete legal sets (stated scenarios); partial set overrides are rejected. The roster replaces usage top-K. --conditions accepts weather, terrain and boolean trick_room. Damage uses the field in both directions, while trick_room changes equal-priority action order. This remains a static reference. See schema §18.

For tune --from-cell, construct {report:<full review>,ref:<member_cell ref>,goal,solve_for, compare_items?,maximize?,probability?}. The tool verifies and recomputes the cell, binds exact sets/actual forms and field, then returns threshold lanes and explicit trial changes. Read each trial's available and qualifications before passing it to try. Unreachable lanes keep their closest percentages; no lane is selected automatically. See schema §19.

speed-ladder reads the complete current-rule team pool. Its reference is the most frequent known speed configuration of each actual form, with exact-form meta marginals as a disclosed fallback. Pass a stated nature/SP/item/ability through --set; an omitted item means no held item, except a Mega's required stone. Read schema.speed_ladder_spec or references/schema.md §10 for field inputs, sample thresholds and limitations. “Reliably/usually” describe the observed sample, not all players.

counters reads the current opponent cache and derives one floor per concrete candidate build. --top-n selects candidate usage ranks (1–60), not a number of recommendations. C2/C1 rows cover all retained builds of the exact target form; conditional_rows keep their C0 floor and name every answered build and blocker. No complete floor is claimed for incomplete_rows. This is a low-confidence static reference, not a win rate or a universal counter. See references/schema.md §11.

ko-scan uses the highest-coverage retained build of the exact named form; an exact variant id selects another build. --direction attacking|defending|both defaults to both. Read the reference, direction and entire qualification group before stating a KO: calculator_guaranteed does not override Sash, Disguise, variable hit counts or other caveats. Missing records are not survival evidence. The KO verdict comes from cached ko_chance, not static percentage buckets. See §12 of the schema.

scenario reads a JSON object with attacker, defender, move, format, optional attacker_set, defender_set and field. It keeps the source build visible, validates stated overrides against the Champions dex, and calculates one stated encounter. The current cache covers retained top-ranked builds; an absent form is an explicit miss. See references/schema.md §13 for the fields and limits.

threshold extends that scenario with solve_for, goal, and optional probability, compare_items, maximize, and speed_field. Each nature/item lane enforces the 66-SP budget; unreachable goals return the closest feasible fact. For two-hit goals, the recovery-aware calculator verdict decides the threshold. Read the selected lane's caveats before stating a battle guarantee. See references/schema.md §14.

Run several operators in one process when they belong to the same request:

python scripts/team.py session session.json

The session spec is a JSON array of operations such as validate, diagnose, select, matchup, and tune. Use the shape emitted by schema. It accepts inline JSON inputs and backward {"$ref":"/0/result/..."} pointers to prior full results. session --spec - reads the array from stdin; --out session-full.json preserves complete results. See references/examples.md for minimal shapes.

Build Gates

Audit intent and, for new builds, drive intake:

python scripts/team.py context-audit --context ctx.json --format json
python scripts/team.py intake --onboarding --context ctx.json --format json
python scripts/team.py intake --next --context ctx.json --answered format,purpose --format json

Ground and evaluate candidates:

python scripts/team.py landscape --game-format double --context ctx.json --format json
python scripts/team.py observed --game-format double --context ctx.json --format json
python scripts/team.py frame --game-format double --context ctx.json --audit-receipt audit.json --format json
python scripts/team.py slate-evaluate slate.json --frame-output frame.json --format json
python scripts/team.py checkpoint slate-output.json --slate slate.json --format json

Audit the answer:

python scripts/team.py draft-init --slate slate.json --slate-output slate-output.json --format json
python scripts/team.py answer-audit draft.json --slate slate.json --slate-output slate-output.json --format json

For build flows, set frame_required:true in the context/slate and preserve matching receipt files. slate-evaluate eliminates invalid or ungrounded candidates and emits the evidence IDs used by the final answer. answer-audit certifies structure and recomputable claims, not strategic quality. Gates start with verdict and next_step. Add --fail-on-verdict to context-audit, slate-evaluate, checkpoint or answer-audit when process exit status must reflect a negative verdict; session also supports this flag for its batch. Summary slate output (--slate-view summary) is display-only; answer-audit needs the full saved result. Unsupported command options are rejected instead of ignored. Use --view summary --out full.json for JSON diagnose, select, oppmatrix, topmeta, frame, tune, intake, matchup and slate-evaluate. The summary discloses omitted evidence; saved files and session references retain full results from the same computation. CHECK floors, missing coverage and qualifications remain visible. topmeta/oppmatrix accept --summary-limit N (default 6) for table display, preserving source order and population ranks. Summary frame/slate cannot certify gates. intake --onboarding --view summary returns only the next numbered batch; carry resolved ids through intake --next --answered .... Saved artifacts keep original receipt-bound inputs.

Real-Team Evidence

The bundled sample library is factual evidence, not a leaderboard:

python scripts/team.py repset Garchomp --game-format double --format json
python scripts/team.py search --game-format double --species Gengar Froslass --format json
python scripts/team.py show <team-id> --game-format double --format json
python scripts/team.py oppmatrix Garchomp --game-format double --vs Charizard --format json
python scripts/team.py topmeta --game-format double --scope season --species Garchomp

topmeta (Top Meta / 竞技环境 / 上位環境) counts the library on the metagame's own fields and returns each entry with both its count and the in-game value. Use it for "what do top-level teams run" and "how does that differ from the in-game data"; name the population and the sample size in the answer, and never treat agreement with repset/frame as corroboration — it is the same library. --scope high is the high-performance slice (a top-8 finish in an event of 16+, or a ladder top-100 rank / 2000+ rating): use it for every high-level wording — 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. The primary standalone query entry is the meta skill's topmeta; this team command remains available for existing sessions. Both reuse the same comparison and report data_granularity:species_aggregate. Read queries[]: unresolved names and unavailable Mega-only distributions remain explicit, without substituting a base-species distribution. Approximate names retain their matching metadata.

Use observed for constraint-shaped retrieval and search/show for exact library inspection. Scope oppmatrix to an attacker; the entire matrix requires an explicit --all and can exceed tens of MB. For high-level team structure, use landscape --game-format double --scope high --format json (or single). This applies Top Meta's performance criteria to native current-rule joint records. Report population.sample_unit and sample_count: these are published library records, not Top Meta's player-deduplicated units. Unknown achievements are excluded, with no broader-population fallback. The ordinary landscape remains the default; --species narrows cores, not all structural distributions. landscape caches structural facts by source content, implementation, options and handover validity. Use --no-cache to recompute; an unavailable local cache never substitutes for missing source data. Adopting an observed team is allowed only with the provenance and adoption-review disclosures required by answer-audit. Never silently reproduce a stored team.

Output And Boundaries

Authored member species, items, abilities, natures and moves accept trilingual dex names. The CLI normalizes a working copy through the public resolver; every fuzzy correction is disclosed, and unresolved/ambiguous values remain for strict validation. Original inputs and receipt-bound objects stay intact.

search, observed, repset, landscape, frame and slate species constraints match the actual battle form: a base with its own dex-declared Mega stone equals that Mega; ordinary forms and Mega X/Y/Z remain distinct. search with both species and its stone is one conjunctive member condition. Unknown form evidence stays unknown beneath boolean NOT. Read identity_gaps and unresolved_conditions; zero matches need not prove absence. observed excludes rows with unresolved forms when an avoid constraint cannot be proved, counting them under excluded_by_identity. Its owned_coverage is a separate inventory count by base species. oppmatrix name selectors read the cached run_form; an exact variant id selects its own build. Frame core grounding uses the same actual species/ability projection; representative sets preserve observed registration values in registration_modes. An ordinary anchor never auto-selects a Mega because the library only contains that form. Re-run legacy frame/slate artifacts when a gate reports receipt_upgrade_required.

  • Use canonical English names in structured JSON; render localized names in the final user-facing response through the dex.
  • Pass --lang zh|ja|en for human-readable CLI output. JSON remains canonical and language-independent.
  • Keep single and double evidence separate. Keep historical regulations explicitly labeled.
  • A six-member roster is a preview toolbox: singles selects 3-of-6 and doubles 4-of-6. Do not judge it as a flat six-member battle lineup; only zero or one brought Mega option can be active per battle.
  • Expose assumptions, confidence, evidence, trade-offs, and omitted alternatives.
  • Do not infer strength from usage, sample performance tags, or provenance tier alone.

References

  • references/assisted_workflow.md: mandatory routing, guided intake, grounding, gates, and final audit.
  • references/schema.md: team-json, provenance, representative-set/cache, build-context, evidence, frame and Top Meta contracts.
  • references/examples.md: inline sessions, slate/frame bindings, draft fields and saved full results.