[dani1005]

What

Facade coverage for what 1.3.0-rc2 (#28) generated, plus one cross-cutting error improvement.

Memory-space purge: purge_request / purge_confirm / purge_get / purge_wait.

• Two calls stay two: there is deliberately no one-call purge, since the nonce step is the safety mechanism.
• The scope ids are only ever what the call names; the client's app_id / project_id defaults never fill them in.
• purge_confirm takes the purge_request result (or a bare nonce). While a purge is queued or running, the API spends the nonce and answers with that job, whatever its scope. Given the request, the facade compares scopes and raises EverOSError (running job on .job) instead of letting another purge pass as the caller's.
• purge_wait shares the task_wait polling loop (backoff, 429/5xx retries); a timeout ends the wait, not the purge.

Owner guard: update(..., user_id=) and delete_by_ids(..., user_id=). A memory owned by anyone else is a 404 on update and skipped on delete. Documented to be set from the caller's own session, never from model output. delete_by_ids returns deleted.

Error codes: EverOSAPIError.code / .message, read from the error body (engine envelope, flat gateway shape, or a plain {"error": "..."}), so callers can branch on purge_in_progress and the like. Unparseable bodies leave both None.

quickstart.md: a purge section, user_id usage, method-table rows.

Checks

• pytest tests test: 274 passed (11 new facade tests, none of the existing ones changed).
• Against the live API, with calls that change nothing: purge_request on an empty project scope (nonce left to expire), extra-id 422, unknown job 404, update / delete_by_ids with user_id on a nonexistent id (404 / deleted=0), and .code parsed from engine and gateway errors.
• Not exercised live: a confirm that collides with a running purge (needs a real purge); covered by unit tests.

🤖 Generated with Claude Code

[gloryfromca]

No blockers; this can merge as far as I am concerned.

Reviewed the complete origin/v1...HEAD diff and surrounding facade/generated API code. I covered repository guidance (none is present), callers and generated model/method contracts, history, backward compatibility of the task_wait polling refactor, test changes (additive; not weakened), and the facade/generated-name architecture constraint. The purge two-step safety, explicit scope handling, owner guards, and structured API error extraction are consistent with the rc2 contract.

Validation: python -m pytest (274 passed), quickstart compile/drift smoke, facade/generated-name smoke, and git diff --check. GitHub CI is also green on Python 3.12, Python 3.14, and build.