Read Domain Memory

Use this before coding when a request uses a domain term or may cross a Context.

1. In an unfamiliar repository, run readiness --repo-root <repo> first. It is read-only and routes only: empty and greenfield need source discovery or Design before any Registry read, and dead needs recovery. Do not call probe as the first command.
2. Run probe --repo-root <repo> --registry-root <root>.
3. Run validate --registry-root <root> --repo-root <repo> and validate the policy.
4. Resolve terms with resolve-terms, then inspect the owner with get-context and get-record.
5. Inspect existing collaboration with analyze-boundary before proposing a new dependency.
6. Before coding, answer these from reviewed records: which boundary keeps each invariant; what must commit together and what may arrive later; what needs identity; which step is a domain decision, which orchestration, which transport; which Context owns each fact and each policy acting on it; what stays true after failure, retry, duplicate, or partial completion; whether a pattern protects a rule or only adds indirection; whether the change leaves the process. Tactical design reasoning explains each. Decide whether a tactical pattern is needed, then choose an idiomatic implementation only after identifying the owner, invariant, consistency boundary, and failure behavior.
7. For a material change, emit the implementation handoff with reviewed facts, forces, unknowns, and proof obligations. Do not invent the decision in Read mode when the change needs Design.

Use reviewed records as constraints. Candidate records are Working Memory: use them to find terms, owners, evidence, and questions to verify, but never present them as settled policy or use them to authorize a material change. Treat stale sources, unverified maps, and missing records as gaps to report. Do not write files in this mode. See the file-backed API.