canonical-memory-verifier

Canonical Memory Verifier is a clean-room, local-first verifier and published skills-only plugin for one narrow external-memory import boundary:

Before an agent relies on imported memory, deterministically prove which claims are current, which exact declared source bytes they reference, and that remembered authority cannot count as live approval.

OpenAI Plugins Directory version 0.1.0 is published at . Directory publication establishes reviewed distribution; it does not make this repository an OpenAI product, endorsement, upstream integration, or security boundary. The repository contains only synthetic data. It does not connect to a server, scan conversations, discover memory directories, write back to a vault, or execute an action described by memory.

Publisher: Sizhe Cheng (GitHub identity). The code is licensed under Apache-2.0. See PUBLISHER.md, LICENSE, and NOTICE.

Five-minute review

Requirements: Python 3.10 or newer. The verifier uses only the Python standard library.

• a complete bundle with one current head per functional key returns

• corrupt graphs, missing evidence, changed source bytes, and ambiguous heads

• remembered authority with approval_eligible: true is rejected whether its

historical time limit has passed or not;

• output omits input paths and wall-clock timestamps, so the same bundle

produces the same digests on different machines.

Run the candidate checks: