# Schema Resolution > Establish the one schema in force: the shape a project declares for its records — the written statements it keeps. Before a record is written, read, promoted, or audited; when a record's place or shape is in question; "which schema", "where does this go", "what kinds exist", "does this conform", or any doubt about the declared shape. - Skill: `ryzhakar/schema-resolution` (Agent Skill) - Install (CLI): `npx skillmds@latest add ryzhakar/schema-resolution` - Raw SKILL.md: https://api.skillmd.com/api/skills/ryzhakar/schema-resolution/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: ryzhakar (https://skillmd.com/u/ryzhakar) - Updated: 2026-09-21 - Page: https://skillmd.com/skills/ryzhakar/schema-resolution --- Resolve a schema for one scope — one project, holding one human owner, one goal set (what the project pursues), one charter (the project's governing file, loaded at every start), one set of records, and one schema; never stretch a scope across two projects. Hold exactly one schema in force over a scope; never hold a second beside it. Follow the chain of pointers starting at the one the charter holds — a pointer being a record whose content is another record's home — reading each target in turn and following any pointer that target holds; never stop on a target that holds a pointer. Take the schema that chain reaches as the schema in force; never read another source once that schema is in hand. Count a missing or unreadable target as a chain reaching no schema; never stop the resolution on one. Read the project's configuration file at `.claude/memento.yaml` when the chain reaches no schema; never look elsewhere for a project's own schema. Take the shipped default — the schema at `config/default.yaml` in this plugin — when the project has no configuration file; never treat a missing configuration file as a missing schema. Take the whole of whichever source answers first; never merge two sources. Let a project's configuration file replace the shipped default outright; never fill a gap in one source from another. Read five declarations from the schema in force: the record kinds a scope admits, each with the one home its records live in and the tier — the storage level — holding them; the event kinds a scope writes down as they happen; the depth of checking a kind's content passes before it stands; the form a record's kind takes — how a record states which kind it is; and the form a record's provenance takes — how a record states where it came from; never read in a shape the schema leaves undeclared. Find the schema's own kind, home, and tier declared beside every other kind; never place the schema outside what it declares. Let the pointer in the charter bring the schema along with it; never copy the schema's content into that file. Take the schema as the owner's own declaration; never author one in the owner's place. Amend the schema on the owner's word; never amend it otherwise. Record what each amendment replaced; never leave it unrecorded. Read a record's kind by the kind form the schema declares; never take a record's kind from anywhere else. Check a record on three points: that its kind is one the scope admits, that it sits at the home that kind gives, and that its provenance matches the declared form; never pass a record on fewer. Leave the event kinds to `event-capture` and the checking depth to `record-promotion`; never apply either declaration here. Count a project's configuration file among a scope's records; never leave one outside them. Take a configuration file's own kind, home, and provenance form from the shipped default; never take them from that configuration file itself. Resolve the schema in force before a configuration file's own check; never hold resolution for that check. Keep the shipped default outside a scope's records; never count it as one. Check the shipped default against nothing; never search for a schema behind it.