← ClaudeAtlas

dnd-checklisted

Validate and audit an existing Dungeons & Dragons 2024 ("5.5") character sheet against a bundled rules catalog and flag every rules error — wrong AC/HP/save DC, over-filled cantrip or prepared-spell slots, a missing granted spell, a skill on the wrong list. Catches 2014/Pathfinder rules applied to a 2024 character, in the sheet and in your own recall: use it even for one number that looks like easy arithmetic ("is 18 AC right?"), because the armour and species values you remember are the 2014 ones. Use for "check my sheet", "is this legal", "something seems wrong with my character", "vérifier ma fiche". Catalog is 2024-only, level-1 only; output in 9 languages. Do NOT use to create a character (dnd-build), look up one rule (dnd-lookup) or rank builds (dnd-optimize).
mlabarrere/dungeons-and-skills · ★ 0 · AI & Automation · score 68
Install: claude install-skill mlabarrere/dungeons-and-skills
# dnd-check — audit an existing sheet (the sheet checker) ## GROUNDING — do not skip **Do NOT trust your training data.** Your training data blends D&D editions (3.5, 5e 2014, 5.5/2024, Pathfinder) into plausible-but-wrong rules; a sheet is arithmetic with citations, so one wrong value makes it illegal. Therefore: - Every rules value comes from the bundled catalog, never from memory (`data/*.json`). - When code execution is available, run `scripts/dnd.mjs` — never compute AC, HP, DCs or spell counts by hand. - Offer only options the resolver returns. - Every value must cite its provenance (source → effect). - If it is not in the catalog, say "Manquant documentaire" — never invent. Read [references/schema.md](references/schema.md) before mapping a sheet to the model, and again whenever you must explain how a computed value was derived — it holds the model shape and every formula. Read [references/grounding.md](references/grounding.md) if the user disputes a finding or asks why the catalog overrides their sheet. ## Accepted input formats The user may hand you their sheet in any shape — you adapt, and you always reply in the user's language (see *Languages* below): - **Free text or markdown:** extract each value (class, species, background, scores, HP, AC, spells…) and assemble the `.character.json` internally. Never ask the user to produce JSON. - **Values dictated one by one:** ask for the next in plain language. - **Missing information:** ask for it one item at a t