メニュー
Build and tune MTG decks across all formats — Commander/EDH, Brawl, Historic Brawl, Standard, Alchemy, Historic, Pioneer, Timeless, Modern, PreModern, Legacy, and Vintage.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Collaboratively build or tune an MTG deck in a live browser UI — the assistant surfaces signal-driven synergy packages, exploration avenues, and ranked candidates (with "why it fits" + honest cost) plus live curve/mana guidance, while you make every decision. Commander family (Commander/Brawl/Historic Brawl), paper + Arena. Reasoning runs in your interactive Claude Code session (no API key, covered by your subscription).
Build and tune MTG cubes — curated card pools designed for drafting. Supports Vintage, Unpowered, Legacy, Modern, Pauper, Peasant, Set, Commander, and PDH cubes.
Generate a Strategy Guide for a finished Commander, Brawl, or Historic Brawl deck — explains how to pilot it, with rules-cited stack/timing/interaction notes. Read-only on the deck; for tuning, run /deck-wizard first.
Search The Gathering Place + Atomic Empire (LGS) and TCGPlayer + Mana Pool (online) for an MTG card list, allocate to minimize total cost across at most three carts, and hand off pre-loaded headed browser windows for checkout.
Render printable PDF proxies for an MTG deck — one PDF for the deck's cards, one for every token kind those cards produce. Cards and tokens both use a layout that mirrors a real MTG card frame (name banner / ASCII art / type banner / oracle text / P/T) with all data pulled live from Scryfall.
Answer MTG rules questions by citing the Comprehensive Rules and Scryfall per-card rulings. Resolves keyword interactions, trigger timing, replacement-effect layer issues, and other card-rules questions with authoritative citations.
| name | deck-wizard |
| description | Build and tune MTG decks across all formats — Commander/EDH, Brawl, Historic Brawl, Standard, Alchemy, Historic, Pioneer, Timeless, Modern, PreModern, Legacy, and Vintage. |
| compatibility | Requires Python 3.12+ and uv. Shares mtg_utils package via symlink. |
| license | 0BSD |
Build MTG decks from scratch or tune existing ones, across every supported format: Commander/EDH, Brawl, Historic Brawl (singleton formats with commanders) and Standard, Alchemy, Historic, Pioneer, Timeless, Modern, PreModern, Legacy, Vintage (60-card constructed formats with sideboards). Every recommendation MUST be grounded in actual card oracle text from Scryfall — never from training data.
The skill has two phases. Phase 1 (Deck Acquisition) determines how the user gets a deck: either by parsing an existing list (Path A) or building from scratch (Path B). Both paths produce a deck JSON + hydrated cache. Phase 2 (Tuning) runs the same 13-step pipeline on that deck regardless of origin.
NEVER assume what a card does. Before referencing any card's abilities, look up its oracle text via the helper scripts. Training data is not oracle text.
Exception (Commander/Brawl only): During commander discovery (recommending commanders to a user who doesn't know what to build), you may use training data to generate a shortlist of candidates. But every recommended commander MUST be verified before presenting — write all candidate names to a JSON list and batch-lookup in one call: scryfall-lookup --batch <candidates.json> --bulk-data <bulk-data-path> --cache-dir <working-dir>/.cache.
For 60-card constructed: discovery (brainstorming candidates) may use training data; evaluation (deciding to include) must use verified oracle text.
| Format | Deck Size | Copy Limit | Sideboard | Platform | Legality Key | Notes |
|---|---|---|---|---|---|---|
| commander | 100 | 1 (singleton) | No | Paper + MTGO | commander | 40 life, multiplayer |
| brawl | 60 | 1 (singleton) | No | Arena | standardbrawl | 25/30 life, Standard pool |
| historic_brawl | 100 | 1 (singleton) | No | Arena (+ paper) | brawl | 25/30 life, all Arena sets |
| standard | 60 | 4 | 15 | Arena + Paper | standard | Rotating |
| alchemy | 60 | 4 | 15 | Arena only | alchemy | Digital mechanics |
| historic | 60 | 4 | 15 | Arena only | historic | All Arena sets |
| timeless | 60 | 4 | 15 | Arena only | timeless | No bans |
| pioneer | 60 | 4 | 15 | Paper + Arena | pioneer | RTR forward |
| modern | 60 | 4 | 15 | Paper + MTGO | modern | 8th Edition forward |
| premodern | 60 | 4 | 15 | Paper + MTGO | premodern | 4th Edition through Scourge |
| legacy | 60 | 4 | 15 | Paper + MTGO | legacy | Eternal, ban list |
| vintage | 60 | 4 (restricted=1) | 15 | Paper + MTGO | vintage | Restricted cards limited to 1 copy |
Before starting, create a TodoWrite list. The first question determines the path; create todos accordingly.
Skeleton Generation expansion: When starting, expand into sub-todos:
Combo-first variant: If the user wants to build around a combo, add/swap:
Skeleton Generation expansion: When starting, expand into sub-todos:
If you draft the whole skeleton in a single batch instead of walking the fill order category-by-category, you still must not batch the sub-todo completions. Either (a) skip creating the fill-order sub-todos entirely for that session (mark only the top-level Skeleton Generation as it starts and completes) or (b) close each sub-todo individually at draft time as you mentally finish that category.
If the user takes the "Outside the Box" workflow, add or swap todos as you reach each alt step, leaving any already-completed standard steps in place.
Do NOT create per-card sub-todos inside any fill step — that's execution detail and would flood the list.
Mark each item in_progress the moment you begin it and completed the moment it finishes — do not batch updates.
Step 6 and Step 8 are long enough that the top-level item alone leaves the user staring at an unchanging list. When you reach them, expand each into sub-todos at that moment (not up front):
Do NOT create per-card sub-todos for the Cut Checklist inside Step 6 — that's execution detail and would flood the list.
Path B adaptations: Steps 9-11 may produce zero changes if the skeleton is sound. If no changes are needed, skip to Step 12.
uv sync --directory <skill-install-dir>
download-bulk --output-dir <skill-install-dir>
Subsequent runs skip if .venv exists and bulk data is fresh (24h).
Use the Write tool for any JSON file containing card names. Apostrophes in names (e.g., "Ashnod's Altar") break shell quoting in heredocs and inline Python. Write tool permissions are cacheable across the session; python3 -c "..." re-prompts every time.
Do NOT write JSON via Bash heredocs (cat > /tmp/foo.json << 'JSONEOF' ... JSONEOF). Heredocs are functionally fine but they produce un-cacheable Bash permission patterns: Claude Code's permission engine bakes the heredoc body into the allow pattern, so every invocation with different content re-prompts the user. The Write tool generates a single Write(/tmp/**) permission that can be granted once and reused.
The same caching trap applies to python3 -c "...", awk '...', jq '...', and any other Bash pattern where the code body varies between invocations. Each unique body is a fresh permission pattern. If you need to extract one field from a JSON file, prefer: (a) passing the file directly to a script that already knows how to parse it, (b) Read with offset/limit on the JSON file, or (c) Grep on the file. Reach for python3 -c only when those options genuinely don't cover the case, and accept the re-prompt cost when you do.
Reuse these stable paths within a session: /tmp/cuts.json, /tmp/adds.json, /tmp/sideboard-cuts.json, /tmp/sideboard-adds.json, /tmp/candidates.json, /tmp/pet-cards.json.
Warning — /tmp files persist across sessions. The Write tool requires a prior Read for any existing file, so the first Write to a reused /tmp/*.json path in a new session fails with "File has not been read yet". If you batched that Write in parallel with a Bash call that reads the same file, the Bash silently consumes the stale prior-session content — the parallel Write error doesn't block the Bash, so cut-check / scryfall-lookup --batch / price-check / build-deck will happily run against the wrong data. Symptom: tool results list cards you never proposed. Rule: (a) at session start, Read each scratch path you plan to reuse before the first Write, OR (b) run Write sequentially (not parallel) and verify success before the dependent Bash call. Never batch Write(/tmp/foo.json) + Bash(tool reading /tmp/foo.json) in a single message. If a tool result lists cards that don't match your intended input, suspect stale-file consumption before suspecting a tool bug.
Alchemy includes two categories of digital-only cards beyond the Standard pool: (1) Rebalanced cards prefixed with A- (e.g., A-Teferi, Time Raveler) that have different oracle text from their paper counterparts, and (2) Digital-only originals with mechanics that only work on Arena (conjure, seek, perpetually, etc.). When building or tuning Alchemy decks, search for both "<Card Name>" and "A-<Card Name>" via scryfall-lookup to verify which version is legal and what its current oracle text says.
AskUserQuestion caps at 4 options — never silently drop the rest. The tool's schema rejects more than 4 options per question. When presenting a longer list (5+ close-call swaps, 5 alternative commanders, etc.), do NOT pick 4 and hide the rest. List every option in the preceding text message with enough detail to decide on, then use AskUserQuestion only as a lightweight picker — either (a) put the top 3 on buttons plus a 4th "Other (specify in notes)" so the user can name any option from the text list, or (b) skip AskUserQuestion entirely for that decision and let the user reply in plain text. The failure mode to avoid is 4 option chips that look like the complete set when the text above mentioned 5+.
uv rebases the working directory to the skill install; relative paths resolve against the wrong root.<working-dir>/.cache, not the skill install directory. Keeping the cache in the working directory avoids outside-workspace permission prompts.scryfall-lookup --batch on the new deck JSON. The hydrated cache is SHA-keyed; old caches go stale silently.Always pass --format <format> to parse-deck once the format is established. Without this, parse-deck defaults to commander and every downstream tool sees the wrong format.
The rarity field in hydrated card data is the default Scryfall printing's rarity, which drifts from Arena's actual wildcard cost. Always use price-check --format <fmt> --bulk-data <path> for Arena wildcard budgeting — it reports the lowest Arena-legal rarity per card by walking every Arena printing.
Some crossover sets use different card names on Arena than in paper/Scryfall: Through the Omenpaths (OM1, the Arena version of Marvel's Spider-Man), Ikoria Godzilla variants, Crimson Vow Dracula variants, and Avatar: The Last Airbender cards all have name discrepancies. Always pass --bulk-data to mark-owned when working with Arena collections — this enables printed_name and flavor_name aliasing so cards like "Skittering Kitten" (Arena name) correctly match "Masked Meower" (Scryfall name). Without --bulk-data, these cards silently appear unowned.
60-card constructed:
{
"format": "pioneer",
"deck_size": 60,
"sideboard_size": 15,
"commanders": [],
"cards": [{"name": "Lightning Bolt", "quantity": 4}, ...],
"sideboard": [{"name": "Roiling Vortex", "quantity": 3}, ...],
"total_cards": 60,
"total_sideboard": 15,
"owned_cards": [{"name": "Lightning Bolt", "quantity": 4}, ...]
}
Constructed formats always have "commanders": []. The sideboard field holds sideboard entries separately from mainboard cards.
Commander/Brawl/Historic Brawl:
{
"format": "commander",
"deck_size": 100,
"commanders": [{"name": "...", "quantity": 1}, ...],
"cards": [{"name": "...", "quantity": 1}, ...],
"total_cards": 100,
"owned_cards": [{"name": "...", "quantity": 1}, ...]
}
All card lists (commanders, cards, sideboard, owned_cards) share the [{name, quantity}] shape. owned_cards starts empty from parse-deck and is populated by mark-owned. price-check reads it to subtract owned copies from the budget; entries with quantity < 1 are treated as "not owned."
Parsed deck JSON is the canonical pipeline intermediate. Once you have a parsed deck JSON from parse-deck, pass it directly to scryfall-lookup --batch and price-check — both scripts accept a parsed deck JSON as <path>, not just a JSON list of name strings. Do NOT extract card names into a separate /tmp/*.json via python3 -c or similar. (The exception is scripts whose input is a subset of the deck — cut-check --cuts <path> and build-deck --cuts <path> --adds <path> expect JSON lists of name strings, not parsed deck JSON, because the caller is specifying which cards to act on. Writing a small /tmp/cuts.json via the Write tool is correct in those cases.)
All scripts are invoked via uv run <script-name> from the skill install directory. Examples in this document elide the uv run prefix for brevity.
After writing or editing a deck text file by hand, always parse it immediately and verify the total card count matches the format's expected size. Off-by-one errors from manual edits are common and silent. Preventive: before writing the deck file, tally your categories visibly in the narrative:
- 100-card (Commander / Historic Brawl): e.g., "Lands 36 + Ramp 10 + Draw 10 + Removal 10 + Wipes 3 + Utility 8 + Engine 18 + Wincons 4 = 99 + 1 commander = 100"
- 60-card singleton (Brawl): scale each category by 0.6 and round. E.g., "Lands 22 + Ramp 6 + Draw 6 + Removal 6 + Wipes 2 + Utility 5 + Engine 11 + Wincons 1 = 59 + 1 commander = 60"
- 60-card constructed: "Creatures 24 + Instants/Sorceries 16 + Lands 20 = 60"
owned_cardsWhen building from a user's collection, use the dedicated helper rather than inline python3 -c:
mark-owned <deck.json> <collection.json> [--bulk-data <bulk-data-path>]
Always pass --bulk-data for Arena collections — this enables printed_name and flavor_name aliasing so crossover cards match correctly. The script is idempotent and safe to chain after every parse-deck / set-commander call.
mark-owned also does front-face aliasing for DFC / split / adventure / modal cards: a deck that lists "Fable of the Mirror-Breaker" (front face only, common in Arena/Moxfield exports) matches a collection entry for "Fable of the Mirror-Breaker // Reflection of Kiki-Jiki" (Scryfall's canonical combined form) and vice versa. You do not need to normalize these by hand before calling mark-owned.
price-check honors deck quantity and owned quantity. Paper (USD) mode charges max(deck_qty - owned_qty, 0) * unit_price per card. Arena wildcard mode applies the same shortfall math plus the Arena 4-cap substitution: owning >=4 of a standard playset-capped card grants effectively infinite supply, but this substitution is suppressed for cards with oracle exemptions (A deck can have any number of cards named X).
| Task | Tool |
|---|---|
| Find format-legal cards by oracle text, type, CMC | card-search --format <fmt> --bulk-data <path> |
| Run the Step-6 deterministic spine (scorecard + candidate swaps) — Commander family | deck-tune <deck.json> <hydrated.json> --bulk-data <path> [--bracket <1-5>] [--max-swaps <N>] [--budget <usd>] --output <wd>/tune.json |
See what the commander/deck cares about (signal lanes) — ad-hoc; the spine's focus has this | deck-signals <deck.json> <hydrated.json> |
Role-density budgets (lands/ramp/draw/interaction/wipes) — ad-hoc; the spine's template has this | slot-budgets <deck.json> <hydrated.json> --deck-size <60|100> |
Rank a separate candidate list by synergy — the spine's swaps already rank adds | deck-rank <deck.json> <hydrated.json> <candidates.json> (from card-search --json) |
| Look up a specific card's oracle text | scryfall-lookup "<Card Name>" |
| View card table (mainboard) | card-summary <hydrated.json> [--nonlands-only] [--lands-only] [--type <T>] |
| View card table (sideboard) | card-summary <hydrated.json> --deck <deck.json> --sideboard |
| Scan the deck for cards matching an oracle pattern | Grep '<regex>' <hydrated.json> — full oracle text, no truncation |
| Count cards / verify total matches deck size | deck-stats <deck.json> <hydrated.json> |
| Find combos in the deck | combo-search <deck.json> --hydrated <hydrated.json> |
| Find combos by card or outcome | combo-discover --card "<Name>" --format <fmt> |
| Check whether a proposed cut breaks a near-miss | Grep '<cut-name>' <combo-search.json> to find near-miss lines where this card appears as a partner |
| Check deck legality | legality-audit <deck.json> <hydrated.json> |
| Check mana base health | mana-audit <deck.json> <hydrated.json> |
| Compare mana before/after | mana-audit <old.json> <old-hyd.json> --compare <new.json> <new-hyd.json> |
| Price check (paper) | price-check <deck.json> --bulk-data <path> |
| Price check (Arena wildcards) | price-check <deck.json> --format <fmt> --bulk-data <path> |
| Apply mainboard + sideboard changes | build-deck <deck.json> <hyd.json> --cuts <c.json> --adds <a.json> --sideboard-cuts <sc.json> --sideboard-adds <sa.json> |
| Compare deck versions | deck-diff <old.json> <new.json> <old-hyd.json> <new-hyd.json> |
| Export for import | export-deck <deck.json> |
| Mark owned cards from collection | mark-owned <deck.json> <collection.csv> [--bulk-data <path>] |
| Know which deck cards I own and how many | mark-owned <deck.json> <collection.json> [--output PATH] [--bulk-data <path>] |
| Plan wildcard spend / get per-card or aggregate Arena rarity | price-check <deck.json> --format <fmt> --bulk-data <path> |
| Get a card's Arena-lowest rarity | price-check --format <fmt> (never the hydrated cache rarity field) |
| Find owned, legal, commander-eligible cards from a collection | find-commanders <collection.json> --format <fmt> --bulk-data <path> --output <working-dir>/.cache/candidates.json |
| Research metagame/strategy | WebSearch + WebFetch (or web-fetch script) |
| Run the self-grill (Step 8 hard gate) | Two parallel Agent calls with subagent_type: "general-purpose" |
| Cite MTG Comprehensive Rules for a ruling | rules-lookup --rule <n> / --term <keyword> / --grep "<regex>" |
| Fetch Scryfall per-card rulings | rulings-lookup --card "<name>" --bulk-data <path> |
| Deeper rules question (layer/timing/stack) | Invoke the rules-lawyer skill via the Skill tool |
Only write python3 -c when none of these cover the need. When you do, batch every related question into a single body — each unique body is a fresh permission pattern, so one big script beats five small ones.
For trivial rules questions that arise during tuning ("what does trample say?", "is this trigger mandatory?"), run rules-lookup --term <keyword> directly — a single CLI call returns the glossary definition plus the relevant CR rule numbers. For nuanced multi-rule questions (layer interactions, replacement-effect timing, stack ordering across triggered and activated abilities), invoke the rules-lawyer skill, which owns the escalation-to-subagent path.
cut-check and legality-audit auto-attach CR citations to their JSON output — --cite-rules is default-on. When a CR file is present next to the deck/hydrated JSON (i.e., you ran download-rules --output-dir <wd> in the same working dir), each flagged interaction / violation carries a rule number + snippet with no extra flag. Pass --no-cite-rules to opt out on a specific invocation; when no CR is reachable the tools still exit 0 and emit a WARN: rule_citations not attached line to stdout.
Run download-rules --output-dir <working-dir> once per session before any Step 1 legality-audit call (24-hour freshness check, same pattern as download-bulk).
Start by asking: "Do you have an existing deck to tune, or do you want to build from scratch?"
If the user provides a deck list (pasted, file path, or URL), follow Path A. If they want to build from scratch, follow Path B.
parse-deck <path> --format <format> --output <working-dir>/deck.json
sideboard field (separate from cards)parse-deck: N cards, M sideboard -> /path/to/deck.jsonAlways pass --format <format> explicitly. Without it, parse-deck defaults to commander and every downstream tool (legality-audit, price-check) sees the wrong format.
Supported formats: commander (100 cards), brawl (60 cards, Standard card pool), historic_brawl (100 cards, Historic/Arena card pool), standard, alchemy, historic, pioneer, timeless, modern, legacy, vintage (all 60 cards with 15-card sideboard). Use --deck-size to override the default deck size.
If the format is not obvious from context, ask the user.
If commanders is empty (common with Moxfield exports that lack //Commander headers), ask the user who the commander is. Don't guess — the first card in the list is often the commander, but not always. Supports partner commanders, friends forever, and background pairings.
Run set-commander <deck.json> 'Commander Name' to move the card from the cards list to the commanders list. set-commander is idempotent — calling it on a card already in the commanders zone is a silent no-op, so it's safe to chain after parse-deck even when the deck file already had a Moxfield Commander header.
If the user has an Arena collection:
mark-owned <deck.json> <collection.csv> --bulk-data <path> to populate owned_cards (accepts both CSV and parsed-deck JSON)mtga-import only for extracting wildcard counts, not collection dataAlways pass --bulk-data for Arena collections (see Tooling Notes > Populating owned_cards for details on aliasing).
scryfall-lookup --batch <deck.json> --bulk-data <path> --cache-dir <working-dir>/.cache
Returns an envelope with cache_path, card_count, missing, digest. The cache file contains all mainboard + sideboard + commander cards hydrated with oracle text, legalities, prices, etc.
Do NOT Read the cache file directly — it's large and floods context. Use card-summary or scryfall-lookup "<Name>" for targeted reads.
Re-hydrate after every deck edit. The hydrated cache path is SHA-keyed against the deck JSON's content, so editing the deck and re-running parse-deck produces a new SHA. If you keep using the old cache_path, downstream tools will show stale data. Any time you modify the deck, immediately re-run scryfall-lookup --batch and switch all downstream script calls to the new cache_path.
After hydration, proceed to Phase 2: Tuning.
Start by determining the format. Ask: "What format do you want to build for?"
If the user says Commander, Brawl, or Historic Brawl, follow the Commander/Brawl Interview below. If they name a 60-card constructed format (Standard, Pioneer, Modern, etc.), follow the 60-Card Constructed Interview.
If the user mentions Arena without specifying a format, clarify:
Each answer informs the next question's options, so ask sequentially via AskUserQuestion:
Note: questions 2-3 can sometimes be merged or skipped (paper is always Bo3; Legacy/Vintage/Modern are always paper).
If the user says "build around a combo" or names a specific combo/build-around card, switch to the combo-first variant:
Combo/Build-Around Discovery:
scryfall-lookup it, then combo-discover --card "<Name>" --format <fmt>combo-discover --result "<outcome>" --format <fmt>combo-discover uses Commander Spellbook, which is crowdsourced primarily by Commander players. Combos that are powerful in 1v1 60-card formats but weak in multiplayer Commander may be underrepresented. Search for "<format> combo decks" and "<card name> combo <format>" to catch format-specific interactions the API might miss.scryfall-lookup --batchAsk: "Do you know what commander you want to build a deck for?"
If yes:
scryfall-lookup to validate it exists.is:commander filter (the source of truth for commander legality — don't try to reimplement the rules).legalities field (key standardbrawl for Brawl, brawl for Historic Brawl).If no: Ask: "Want to explore standard archetypes or go outside the box with unusual combos?"
If the user provides a collection export (Untapped.gg CSV, Moxfield CSV, or any deck-list export) of cards they own and wants to pick a commander from what they already have, follow this exact procedure. Do NOT write ad-hoc Python that loads the file and calls .get() on it. Note: collection exports often contain entries with quantity 0 (tracked/wishlisted cards the user doesn't own). parse-deck preserves these quantities and mark-owned / find-commanders filter by --min-quantity 1 by default, so quantity-zero entries are excluded automatically.
This narrows the candidate pool; it does NOT replace the guided interview. Run the interview anyway — the answers drive ranking against the narrowed pool.
Workflow:
Arena players — ask for a collection CSV first. If the user mentions Arena, Brawl, Historic Brawl, or wildcards, ask: "Do you have a collection export from Untapped.gg, Moxfield, or a similar tracker? A CSV with card names and quantities is the most reliable way to know what you own." If they have one, proceed to step 1. If they don't have a collection CSV, fall back to mtga-import as a last resort — but warn the user: mtga-import reconstructs a collection from saved Arena decks, which is unreliable because Arena allows building decks with unowned cards. Always run mtga-import for wildcard extraction regardless — it reads InventoryInfo from Player.log which is reliable for wildcard counts: mtga-import --bulk-data <bulk-data-path> --output-dir <working-dir>. This writes wildcards.json. Linux users need to pass --log-path explicitly.
Parse the collection — parse-deck <absolute-path-to-collection.csv> produces a parsed deck JSON. parse-deck handles Untapped.gg CSV, Moxfield CSV, Moxfield deck export, Arena, MTGO, and plain text.
Find commander candidates — find-commanders <parsed.json> --bulk-data <bulk-data-path> --format <format> --output <working-dir>/.cache/candidates.json [--color-identity <ci>] [--min-quantity 1]. Always pass --output — the default path is under $TMPDIR which triggers outside-workspace permission prompts. Stdout is a compact text table with columns EDHREC rank, color identity, CMC, name, type_line, flags (PARTNER / BACKGROUND / GC). The last line is Full JSON: <path>.
Run the guided interview (colors, playstyle, mechanics, favorite cards, play group, bracket, budget) the same as the no-collection flow. The candidate pool is the constraint; the interview answers are what differentiates one commander from another.
Build a mixed shortlist of ~5 candidates, weighted by interview answers. Do NOT just pick the 5 lowest edhrec_rank values — that produces boring recommendations. Aim for roughly:
edhrec_rank) that obviously fit the user's stated preferences. These exist to give the user a safe pick.edhrec_rank whose oracle text mechanically matches the interview answers (especially the mechanics question). These are usually the most interesting options.game_changer flag and your judgment about combo density. Do NOT use any "EDHREC bracket" field — community bracket data is user-reported and unreliable.Enumerate partner pairings from within the owned pool. Walk the candidate list once: for each card with is_partner=true, find compatible partners from the same list. For each card with has_background_clause=true, find Backgrounds in the candidate list. Surface promising pairings as wildcard or off-meta picks.
Present the shortlist following the "Commander Recommendation" rules below. Mention that candidates are filtered to cards they own.
Ask: "What format are you building for?"
Arena naming confusion: On MTG Arena, "Brawl" (the queue name) actually refers to Historic Brawl, and "Standard Brawl" refers to what we call Brawl. If a user says "I play Brawl on Arena," they almost certainly mean Historic Brawl. Clarify which they mean.
If Brawl or Historic Brawl: ask "Are you playing on Arena or in paper?"
If Brawl: any legendary planeswalker can be your commander (not just those with "can be your commander" text). Vehicles and Spacecraft with power/toughness are also eligible in all formats.
Colorless commanders in Brawl: If the chosen commander has no colors in its color identity, note that the deck may include any number of basic lands of one chosen basic land type.
Partner mechanics are available in all formats.
Colors — "What colors do you enjoy playing? (Pick any combination, or 'no preference')"
Playstyle — "What's your preferred playstyle?" Present options with brief plain-language explanations so newer players can follow:
Mechanics — "Any specific mechanics you enjoy?" Offer examples with explanations: "+1/+1 counters (growing your creatures over time), theft (stealing opponents' cards), blink (flickering creatures to reuse their effects), spellslinger (casting lots of instants and sorceries), artifacts-matter, landfall (rewards for playing lands)." Open-ended. If the answer maps to multiple distinct sub-archetypes, ask one follow-up to disambiguate.
Favorite cards/sets — "Any favorite cards or recent sets that excited you? This helps me find commanders in a similar design space."
Play group dynamics — "How does your play group typically play? (casual/competitive, combo-heavy, creature-heavy, lots of interaction)"
Bracket — "What power bracket are you targeting? (1-4, or casual/mid/high/max)"
Budget — "What's your total budget for the deck? (dollar amount, or wildcard counts for Arena)"
After the guided interview, recommend 3-5 commanders that fit. Partner pairs and commander + background pairings are valid recommendations — present each pair as a single option with their combined color identity. You may use training data to generate a shortlist — this is commander discovery, not card evaluation, so the Iron Rule does not apply at this stage. However, every recommended commander MUST be verified via scryfall-lookup before presenting.
Present each recommendation with:
Let the user pick.
Ask all of these (skipping any already answered during the guided interview):
For pet cards: write all pet card names to a JSON list and batch-lookup in one call: scryfall-lookup --batch <pet-cards.json> --bulk-data <bulk-data-path> --cache-dir <working-dir>/.cache. Verify each exists and is within the commander's color identity. Slot pet cards into the appropriate template categories — they count against those category budgets. If pet cards exceed ~10, warn the user that it limits the ability to build a balanced skeleton.
After the interview, check whether the deck's constraints naturally fit a Companion. Companions are powerful (a guaranteed extra card) and should be actively considered:
card-search --format <fmt> --bulk-data <path> --oracle "Companion" --type "Creature" to find format-legal Companionsparse-deck --deck-size 80 if selectedInfer from answers. If unclear, ask directly: beginner, intermediate, or advanced?
| Level | Interview Depth | Explanation Style |
|---|---|---|
| Beginner | Explain format basics, define archetypes | Full sentences, analogies |
| Intermediate | Assume format knowledge | Concise, note key interactions |
| Advanced | Assume deep knowledge | Tables, shorthand, focus on novel angles |
If the user provides pet cards:
scryfall-lookup --batch to verify existence and oracle textlegalities.<format> field)"<format> metagame 2026" or "<format> tier list" to find current top archetypesweb-fetch for bot-blocked sites) the top result to get archetype names and metagame share percentages"<archetype name> <format> decklist 2026" to find a sample listweb-fetch script as fallback if WebFetch is blocked (browser headers + curl fallback).parse-deck --format <fmt> <path> --output <working-dir>/deck.jsonscryfall-lookup --batch <deck.json> --bulk-data <path> --cache-dir <working-dir>/.cacheFallback if no sample list is obtainable:
card-search queriescard-search --format <fmt> --bulk-data <path> --oracle "<keyword>" --color-identity <colors> to find build-around cardscombo-discover for combo shells in those colors"<format> <archetype/mechanic> deck" for inspirationcard-search)Scryfall lookup — Run: scryfall-lookup "<Commander Name>"
Read the full oracle text, color identity, CMC, and types. For partner/background pairs, look up both commanders and note how they interact with each other.
EDHREC research — Run: edhrec-lookup "<Commander Name>"
For partner commanders: edhrec-lookup "<Commander 1>" "<Commander 2>"
Review top cards, high synergy cards, and themes. Brawl/Arena note: EDHREC data is sourced from Commander/EDH decks. For Brawl/Historic Brawl, EDHREC recommendations must be legality-checked against the deck's format before including. For Arena decks, also verify cards exist on Arena.
Web research — Use WebSearch for the commander + "deck tech", "strategy", "guide". Use WebFetch or the helper script to read strategy articles:
Run: web-fetch "<url>" --max-length 10000
Strategy synthesis — Summarize the commander's key mechanics, primary strategies, and synergy axes. For partner/background pairs, identify how both commanders contribute to the strategy and where their mechanics overlap or complement each other. Present to the user for validation. If the user defers or has no preference, default to the commander's most popular theme on EDHREC and move forward.
The goal is building enough understanding to make smart category fills — not deep analysis (Phase 2 handles that).
If EDHREC has no data for the commander (new or obscure cards), fall back to:
card-search to find cards that mechanically synergize with the commander's keywords/oracle text within the commander's color identity. For Arena decks, use --arena-only. For paper Brawl, use --paper-only to exclude Arena-only digital cards.Ask (accepting either or both):
card-search oracle patterns and combo-discover --card queries)combo-discover --result query, e.g., "infinite tokens", "infinite mana", "mill entire library")Two discovery strategies based on user input:
card-search with oracle text to find cards matching mechanics, then feed card names to combo-discover --card "X" --card "Y"combo-discover --result "Infinite X" directly--result and --card filtersObscurity to popularity mapping:
--sort -popularity, skip top results, popularity 1000-10000--sort popularity, skip 0s, popularity 100-1000--sort popularity, popularity 0-100Jankiness to max combo size:
For commander known + outside the box: use combo-discover --color-identity <commander-CI> to constrain to the commander's colors.
Before presenting, write all combo piece names across all combos to a JSON list and batch-lookup in one call: scryfall-lookup --batch <combo-pieces.json> --bulk-data <bulk-data-path> --cache-dir <working-dir>/.cache.
Present 3-5 interesting combos with:
Ask: "Want to build around one of these, or combine multiple?" If combining, verify color identities are compatible.
Two-wave search for each selected combo:
card-search --is-commander --color-identity <combo-CI> --oracle "<combo-keyword>" — commanders whose oracle text mentions the combo's mechanicsscryfall-lookup --batch <fit-candidates.json> --bulk-data <bulk-data-path> --cache-dir <working-dir>/.cache (Iron Rule applies).Also check if any combo piece IS a legendary creature that could be the commander.
Present 2-3 commander options with:
User picks a combo + commander pairing.
| Category | Typical Count | Notes |
|---|---|---|
| Threats/Creatures | 12-28 | Varies hugely by archetype (aggro: 28, control: 4-8) |
| Removal/Interaction | 4-12 | Format-dependent; more in slower formats |
| Card Advantage | 4-8 | Draw spells, planeswalkers, selection |
| Utility/Flex | 2-8 | Archetype-specific support |
| Lands | 20-27 | Constructed land formula (see below) |
These counts are guidelines, not rules. Aggressive decks skew toward threats; control decks skew toward interaction and card advantage. The archetype defines the balance.
Use the constructed land target formula:
Run mana-audit after building to verify.
| Land Type | Budget | Mid-Range | High-End |
|---|---|---|---|
| Basics | 10-16 | 6-10 | 2-6 |
| Dual lands (shock, fast, check) | 4-8 | 8-12 | 8-12 |
| Fetch lands | 0 | 0-4 | 8-12 |
| Utility lands | 0-2 | 2-4 | 2-4 |
For Arena formats: Khans fetchlands are available (Pioneer and up), but Zendikar/Onslaught fetches are not. Use the Arena-available mana base (shock lands, fast lands, pathway lands, triomes, Khans fetches for Pioneer/Historic/Timeless).
The land count formula gives a total, but the color mix matters just as much. Use mana-audit output (pip demand % vs. land production %) to verify, and follow these rules of thumb:
mana-audit flags a color deficit >5%, replace basics with duals that produce the deficient color.For each category:
card-search --format <fmt> --bulk-data <path> --oracle "<pattern>" --color-identity <colors> for candidatesscryfall-lookup --batch to verify oracle text of all candidatesThe sideboard is 15 cards designed to improve matchups after Game 1.
Process:
card-search --format <fmt> --bulk-data <path> --oracle "exile.*graveyard"card-search --format <fmt> --bulk-data <path> --oracle "destroy.*artifact"scryfall-lookupSideboard card counts:
card-search to find: tutors that find combo pieces, protection for combo assembly, redundant effectsCard counts scale with deck size. The base counts below are for 100-card decks; for 60-card decks, multiply each count by 0.6 and round to the nearest integer.
| Category | 100-card | 60-card | Notes |
|---|---|---|---|
| Commander(s) | 1-2 | 1-2 | Already selected |
| Lands | 36-38 | 22-23 | Burgess formula scaled: round((31 + colors + cmc) * deck_size / 100) |
| Ramp | 10 | 6 | Mana rocks, dorks, land-fetch spells |
| Card draw | 10 | 6 | Prefer draw that aligns with strategy |
| Targeted removal/disruption | 5-12 | 3-7 | Scaled to bracket |
| Board wipes | 2-5 | 1-3 | Scaled to bracket |
| Win conditions | 3-5 | 2-3 | Cards that close out a game |
| Engine/synergy pieces | 15-20 | 9-12 | Cards that work with the commander |
| Protection/utility | 8-10 | 5-6 | Counterspells, hexproof, recursion |
The category counts above are defaults — adjust them after strategy validation to match the user's confirmed direction. Examples:
Hard constraints that don't flex: Lands and ramp stay at Burgess formula minimums regardless of strategy. Total card count must match the deck's expected size (100 for Commander/Historic Brawl, 60 for Brawl, or the user's specified size).
The land count comes from the Burgess formula, but composition matters. Guidelines:
Run mana-audit after filling to verify color balance.
Based on Command Zone #658 (2025), EDHREC, and MTGGoldfish guidelines:
| Category | Bracket 1-2 (Casual) | Bracket 3 (Upgraded) | Bracket 4 (Optimized) |
|---|---|---|---|
| Targeted removal/disruption | 5-7 | 8-10 | 10-12 |
| Board wipes | 2-3 | 3-4 | 4-5 |
| Total interaction | 8-10 | 12-14 | 15-18 |
"Disruption" includes counterspells, discard, and stax pieces — not just creature/artifact removal. Extra interaction slots come out of the engine/synergy budget.
Wildcard-constrained path (Arena, under 10 combined rare + mythic wildcards): When the user's rare + mythic wildcard budget totals less than 10, invert the fill strategy. Instead of drafting ideal cards and checking ownership after, search the user's collection for owned cards that fit each category first. Only spend wildcard slots on commons and uncommons that fill critical gaps.
Category fill order matters. Fill foundational categories first:
Per category:
card-search to find synergistic cards EDHREC may not surface. For Arena decks, use --arena-only and omit --price-max. For paper Brawl, use --paper-only to exclude Arena-only digital cards.scryfall-lookup --batch <candidates.json> --bulk-data <bulk-data-path> --cache-dir <working-dir>/.cache. Run card-summary <cache_path> to see a compact table. Only Read the full cache file for the ~3-5 candidates you're actually deciding between, using offset/limit.card-search to find supporting cards: tutors that find combo pieces, protection for combo pieces, redundant effectsRun all four checks in order. If any fail, fix and re-check from the top. Re-hydrate after any deck edit.
legality-audit <deck.json> <hydrated.json>
Checks: format legality, copy limits (singleton for Commander/Brawl; 4-of rule + Vintage restricted for constructed), sideboard size (max 15 for constructed), deck minimum. Fix any violations before proceeding.
Commander/Brawl note: Historic Brawl bans many Commander staples (Sol Ring, Skullclamp, Hour of Reckoning, Triumph of the Hordes, etc.) that look like obvious includes if you're thinking in Commander terms — this check catches them. Color-identity violations typically mean a card slipped past the commander's identity gate.
# Paper
price-check <deck.json> --budget <budget> --bulk-data <path>
# Arena
price-check <deck.json> --format <fmt> --bulk-data <path>
If over budget, substitute expensive cards with budget alternatives. Re-run after changes. For Arena, "most expensive" means highest rarity — swap rare cards for uncommon alternatives. Watch the illegal_or_missing warning line for cards that escaped the legality audit.
deck-stats <deck.json> <hydrated.json>
Verify: total card count matches expected size, land count matches formula target, category distribution looks reasonable, sideboard count (0-15 for constructed).
mana-audit <deck.json> <hydrated.json>
Uses the Burgess/Karsten formula for Commander/Brawl, constructed land formula for 60-card formats. Checks land count and color balance. FAIL means the mana base needs fixing before proceeding.
This is a gate — do not present a skeleton that fails any of these checks. If any check fails and you edit the deck to fix it, re-parse, re-run scryfall-lookup --batch to refresh the hydrated cache, and re-run ALL checks from the top.
Present as a categorized list with brief synergy notes:
## Creatures (24)
4 Monastery Swiftspear — hasty one-drop, prowess triggers off spells
4 Eidolon of the Great Revel — punishes low-curve opponents
...
## Instants/Sorceries (16)
4 Lightning Bolt — premium removal + face damage
...
## Lands (20)
8 Mountain
4 Inspiring Vantage — untapped on turns 1-3
...
Sideboard (Bo3 only):
## Sideboard (15)
3 Roiling Vortex — vs. free spells, lifegain decks
2 Rampaging Ferocidon — vs. token/lifegain strategies
3 Smash to Smithereens — vs. artifact-heavy decks
...
Present organized by the builder's categories (lands, ramp, card draw, removal, board wipes, win conditions, engine/synergy, protection/utility). Include brief notes on why key synergy cards were included.
Mainboard: 60 cards | Lands: 20 | Avg CMC: 1.85
Sideboard: 15 cards
Budget: $45.20 / $50.00 (paper) or 2M/4R/8U/12C (Arena wildcards)
For Commander/Brawl, show:
For Arena decks, present a wildcard cost table:
Rarity Skeleton Cost Budget Remaining Mythic X Y Z Rare X Y Z Uncommon X Y Z Common X Y Z
Present the skeleton and ask: "Want to make any adjustments before we move on to tuning?"
If the user requests changes, apply them, re-run structural verification, and present again.
<working-dir>/<deck-name>.jsonexport-deck <deck.json> -> <working-dir>/<deck-name>-moxfield.txt<working-dir>/.cache/hydrated-<sha>.jsonAfter presenting and getting user approval, proceed to Phase 2: Tuning.
Both paths converge here. The deck JSON and hydrated cache are ready.
Run in this order (cheapest-to-fail first):
legality-audit <deck.json> <hydrated.json>
Checks: format legality, copy limits (singleton for commander formats, 4-of + Vintage restricted for 60-card), sideboard size, deck minimum, color identity (commander formats). Must PASS before continuing. If FAIL, surface violations and ask user how to fix.
deck-stats <deck.json> <hydrated.json>
Review: total cards, land count, creature count, ramp count, avg CMC, curve distribution, sideboard total (60-card). Note any obvious red flags. Flag immediately if the total card count does not match the deck's expected size.
Review the alternative_cost_cards section. For any card with alternative costs (suspend, adventure, foretell, etc.), note the cost most likely to be used in this deck. Do not evaluate these cards at their CMC alone.
card-summary <hydrated.json> --nonlands-only
card-summary <hydrated.json> --lands-only
card-summary <hydrated.json> --deck <deck.json> --sideboard # 60-card only
Scan mainboard (and sideboard for 60-card) oracle text. Flag cards with alternative costs for adjusted CMC evaluation.
mana-audit <deck.json> <hydrated.json>
Notes land count status (PASS/WARN/FAIL) and color balance. Uses the Burgess/Karsten formula for commander formats and the constructed formula for 60-card formats.
Check the sideboard for a Companion card (card-summary <hydrated.json> --deck <deck.json> --sideboard and look for the Companion keyword). If one exists, note its deck-building restriction — all proposed changes must continue to meet it.
If no Companion exists, check whether the deck naturally meets one's restriction. Companions are powerful enough that a deck accidentally qualifying for one (e.g., a low-curve aggro deck meeting Lurrus's "no permanents with mana value > 2") should actively consider adding it. Use card-search --format <fmt> --bulk-data <path> --oracle "Companion" --type "Creature" to find candidates, then check restrictions against the current deck. If one fits, suggest it in Step 6 as an addition (it takes 1 sideboard slot).
Ask via AskUserQuestion:
Format-specific context to mention when relevant:
For Arena, use compact notation: NM/NR/NU/NC (e.g., 2M/4R/8U/12C = 2 mythic, 4 rare, 8 uncommon, 12 common wildcards).
If the user ran mtga-import, check for wildcards.json in the working directory before asking about budget.
Confirm carry-forward context from Phase 1: format, platform, Bo1/Bo3 (60-card) or bracket (commander), budget (total/spent/remaining), experience level, archetype, Companion (if any), max swaps. Don't re-ask questions already answered.
When arriving from Path B, note both the total budget and the upgrade budget (total minus skeleton cost). Use the upgrade budget for swap decisions during analysis. Track owned cards separately — they don't count toward either budget.
Path B default: Suggest max swaps of 15-20 for a fresh skeleton (user can adjust). Pain points default to "general optimization" unless the user raised specific concerns during Phase 1.
"<format> metagame 2026" or "<format> tier list""<deck archetype> <format> sideboard guide" and "<deck archetype> <format> matchup guide"Research was already done in Phase 1. Verify metagame alignment rather than repeating:
Run: edhrec-lookup "<Commander Name>"
For partner commanders: edhrec-lookup "<Commander 1>" "<Commander 2>"
Also use WebSearch for the commander + "deck tech", "strategy", "guide" to find Command Zone, MTGGoldfish, and other content creator analysis.
Fetching strategy articles: Use WebFetch first. If it returns an empty JS shell or navigation-only content, fall back to the helper script:
Run: web-fetch "<url>" --max-length 10000
Key principle: Research informs but doesn't dictate. EDHREC popularity doesn't automatically make a card right, and unpopularity doesn't make it wrong.
Combo discovery (optional): For deeper exploration beyond what combo-search surfaces, use combo-discover to find combo lines the deck could support. Search by cards already in the deck (--card "Key Card") or by desired outcomes (--result "Infinite X").
Brawl format note: EDHREC data is sourced from Commander/EDH decks. When tuning a Brawl deck, EDHREC recommendations must be legality-checked against the deck's format before recommending. Use card-search --format <format> --bulk-data <path> to verify candidates are legal. For Arena decks, also use --arena-only.
Commander analysis was already done in Phase 1. Verify alignment rather than repeating:
combo-search <deck.json> --hydrated <hydrated.json>
Surface existing combos and near-misses. Note which near-misses could be completed with 1-card additions.
Before recommending any near-miss as an add, Read the full description field for that near-miss from the combo-search JSON (use offset/limit). The compact text report shows only the card list and the result; the JSON's description spells out the activation sequence and frequently names additional required pieces not in the card list. Recommending a near-miss without reading its description is an Iron Rule violation.
Supplement with WebSearch (60-card): combo-search uses Commander Spellbook, which is crowdsourced primarily by Commander players. Combos that are powerful in 1v1 60-card formats but weak in multiplayer Commander may be underrepresented or missing. Search for "<archetype> combo <format>" and "<key card> combo <format>" to catch format-specific interactions the API might miss.
Bracket compliance (Commander): Check combo results against the user's target bracket:
Present your understanding of the deck's:
cut-check and determines how trigger values are evaluated.Ask the user to validate or correct. This alignment prevents wasted analysis on swaps that fight the deck's identity.
Skip this step entirely for 60-card constructed formats. Mark it completed immediately and proceed to Step 6.
Before evaluating individual cards, systematically check for mechanical interactions between the commander and every card in the deck. This step catches synergies that are invisible when reading cards in isolation.
List the commander's keywords and any keywords granted by cards in the deck. Check every pair for emergent effects:
This applies to all cards, not just the commander. Equipment and auras that grant keywords to the commander are especially important.
Identify the commander's core multiplier (extra upkeeps, extra combats, extra turns, extra phases, trigger copying, token doubling, etc.). For EVERY triggered ability in the deck that fires during the multiplied window, calculate its output at 1x, 3x, and 5x the base rate. Present and evaluate the multiplied value, not the base value.
A "1 damage to each opponent" trigger looks marginal at 1x. At 5x with 3 opponents, it's 15 damage + 15 life — a legitimate win condition. Evaluate accordingly.
For commanders whose trigger scales with combat damage dealt, explicitly identify pump as a strategic pillar. Each +1 power is not just +1 damage — it's +1 trigger of every effect in the multiplied window.
For each card, ask: "Does this card's output feed back into its own input or the commander's trigger condition?" Examples:
Cards with feedback loops are almost always stronger than they appear in isolation. Flag them before the analysis phase.
Identify all cards that return themselves to a usable zone: re-suspend, buyback, retrace, escape, flashback, "return to hand" clauses, "exile with time counters" effects. Evaluate these on their per-game value (total free casts over a typical game), not their per-cast value. A 6-mana spell that re-suspends and gets cast for free every 1-2 turns is a permanent with a triggered ability, not a one-shot.
Identify cards that multiply the commander's impact. Two categories:
Commander copies — cards that create token copies or become copies of the commander (Helm of the Host, Spark Double, Clone effects, Mirror March, Followed Footsteps). Non-legendary copies bypass the legend rule and retain all triggered and activated abilities.
Ability copiers and trigger multipliers — cards that copy or double the commander's triggered/activated abilities (Strionic Resonator, Rings of Brighthearth, Panharmonicon for ETB commanders, Teysa Karlov for death triggers, Isshin for attack triggers, Seedborn Muse for extra activations).
Scan oracle text for these patterns directly — cut-check's commander_multiplication field catches obvious cases but misses oddly-worded effects. These cards are force-multipliers. Treat any card you flag as untouchable when drafting cuts — it should not appear on the cuts list without explicit justification.
Run: combo-search <parsed-deck-json> --output /tmp/combo-search.json
Review existing combos and near-misses. Distinguish:
Commander / Brawl / Historic Brawl — run the deterministic spine first. One deck-tune call replaces the mechanical counting and drafting this step used to do by hand. It runs the same deterministic tuner deck-forge uses (ADR-0023/0029) and returns a scorecard + candidate swaps:
deck-tune <deck.json> <hydrated.json> --bulk-data <path> \
--bracket <1-5> --max-swaps <N> [--budget <usd>] [--shape <aggro|midrange|control|combo>] \
--output <working-dir>/tune.json
Scorecard sections and what each subsumes:
shape / efficiency — deck speed + curve/tempo health (the 6a curve read).template — role density (lands / ramp / draw / interaction / wipes) vs the Command-Zone bands (the old slot-budgets pass; 6a + 6b counts).focus — the commander's signal lanes and whether the deck concentrates on them (the old deck-signals pass; 6c).mana — the full mana audit: color balance, Burgess land target, untapped quality.curve — the per-CMC histogram.combos — combos + near-misses; combo pieces are auto-protected from the proposed cuts.bracket — the constraint gate when you pass --bracket (Game Changers / mass land denial / extra turns / two-card combos vs the target bracket's official allowances; ADR-0030).swaps — budgeted (cut, add) candidate pairs, synergy-ranked (the old deck-rank pass).Your job is judgment, not counting. Read the scorecard's top_issues, then run the per-card Cut Checklist (6d) on the candidate swaps deck-tune surfaced — verify oracle text, commander interaction, multiplied trigger values (Step 7 cut-check), and combo lines before accepting any swap. The Self-Grill (Step 8) still gates the final proposal, and you and the user make every final call (the user drives card choices).
What deliberately stays OUTSIDE the spine:
mana-audit --compare <old> <new> for the before/after check in 6g / Step 10 (the scorecard carries the current mana read; the comparison is its own call).bracket gate governs permission (Game Changers etc.), not interaction density.archetype-audit is now optional — focus answers the commander-coherence question. Reach for archetype-audit only to test a specific named theme's density or to find bridge cards (capabilities focus doesn't provide).60-card constructed: deck-tune refuses non-commander formats (the tuner is commander-shaped). Use the existing agent-driven analysis below — metagame archetypes, build-around evaluation, sideboard.
mana-audit for quantitative checkRole-density budgets (deterministic). Commander family: the spine's scorecard template section already has the count of each role (lands / ramp / card_draw / interaction / board_wipe) against the Command-Zone band — read it there, don't re-run a separate pass. A role showing (under)/(over) is a falsifiable signal of where to add or cut. (The standalone slot-budgets <deck.json> <hydrated.json> --deck-size <60|100> [--shape ...] CLI remains for ad-hoc use, e.g. 60-card constructed, but it's the same budgeter the scorecard runs.)