Open source · MIT
Three of 46 Legend breakdowns survived the audit unchanged
This is an open-source knowledge base that turns a general AI agent into a Riftbound TCG specialist — a Claude Code skill, MIT-licensed — together with the record of where it was wrong. Every Legend entry in it was derived from card text and game mechanics rather than summarised from other people’s guides. That makes the entries cheap to regenerate and, on their own, unproven, so all 46 were checked against established play. Three needed no changes. The rest each had at least one thing to fix, most often a default-Champion assumption that ran backwards or an archetype label real coverage does not use.
View on GitHub ↗What checking it against real play changed
The log is in the repository, one row per check, and its most useful row is not a Legend at all but a failure in the ban-list substitution routine. That routine reads what a banned card does inside a specific deck, then looks for a legal card doing the same job. A ramp build of Master Yi lost Obelisk of Power; the routine offered Startipped Peak and Fortified Position — same domain, same currency, entirely legal.
What it dropped was the condition. Obelisk of Power hands a rune to every player on their own first turn, unconditionally. Both replacements pay out only to whoever holds or defends the point, and this build concedes the early board on purpose in order to ramp — so the swap would have funded the opponent. Asking instead who collects turns up a different answer entirely: the cards that suit this deck tax the player who does contest the point, and a search for “cards that give runes” cannot reach them. The same swap into a differently built Master Yi deck would have been right, and two creator builds of this Legend that week sat on opposite sides of it. A player outside the project found this, not an internal check.
Auditing the extraction step turned up a second, quieter class of error: the deduplication keyed on domain alone, so wherever a Legend had a third champion print sharing a domain with a known one, that print was never shown to the derivation at all. At least 8 Legends were affected. Three surfaced during ordinary research, because the omitted print was the one real guides could not avoid mentioning; the other five came out only by reading the extraction script against the card data, since their omitted prints are off-meta and never come up in coverage. Both corrections went into the process rather than into the individual entries, on the rule that a mistake which repeats is a defect in the method.
The shape of the result is consistent across all 46: derivation from card faces holds up, and the moment the question becomes what a particular deck is actually trying to do, it needs a human.
What it is
A Claude Code skill: structured knowledge that lets an AI agent reason about a trading card game the way a serious player or a guide editor does, rather than look rules up. How to build around a Legend’s ability instead of stacking strong cards; how to see that an imported list will not survive the current ban list before anyone sleeves it; how chain and priority actually resolve inside a single showdown.
It runs in three modes with stated boundaries. Deck coach gives recommendations. Rules consult gives cited explanations. Play decision proposes candidate actions, and every one of them needs a human to confirm it. It does not adjudicate rules, does not change game state on its own, and is not built for use at an event.
It was written to produce this site’s deck guides. The public version has everything specific to this site’s deployment removed, leaving the part that stands alone.
What is in it
Deckbuilding
Archetype theory, curve and rune ratios, a build order that starts from the Legend, hypergeometric draw tables computed rather than estimated — over 39 cards, not the 40 of the main deck, because the Champion leaves before the shuffle — and the routine above for deciding whether a well-known list is still playable and what to swap in. One document generalises what a new market looks like on day one: the card pool is not merely smaller, it also inherits a ban list calibrated against cards that pool cannot legally contain yet.
Play procedure
The six-phase turn, the chain / priority / focus system, step-by-step resolution of combat and showdowns, a mulligan framework, and a layered method for piloting one specific list. Sourced from the official Core Rules plus cross-checked community explanations, cited throughout.
Rules consultation
Answers to “how does this interaction actually resolve”, each carrying its citations, a source tier and a confidence level. Tournament-level disputes are returned as “ask a judge” rather than answered. It is an unofficial reading, and every conclusion can be walked back through its references.
Alongside those, a roster of all 46 Legends, each written independently and tagged with archetype, win route and searchable mechanics. Deriving them from card text rather than from published guides is what makes them cheap to regenerate and keeps them from expiring with the current best list — and it is what put them through the audit above.
What it does not claim
Anything with an expiry date — ban lists, errata — is looked up at query time and never baked into a prompt. The card data bundled with the repository is a snapshot taken on 2026-08-16 and is not guaranteed to be reproducible from scratch. The derivations are hypotheses; the audit log is the only thing that makes any of them more than that, and it covers Legend breakdowns, not the whole knowledge base.
Where it runs
Riftbound Chronicle is its first production user. 7 of the deck guides on the site have their play analysis generated from this knowledge base — each carries an “AI analysis, for reference” badge and a link to the methodology, and the decklists themselves come from overseas sources. The remaining guides are translations of overseas coverage or written by hand, labelled as such. Problems hit while writing guides go back into the skill.
The repository also ships three install-free demo pages, one per mode, and an offline bridge that turns a pasted Rift Atlas deck URL into deck-coach input without scraping anything. Licence is MIT. It is not affiliated with Riot Games.