Shadow backfill tools for TencentDB Agent Memory historical conversations

[leoge007]

I built a small shadow-only toolchain for safely backfilling historical agent conversations into TencentDB Agent Memory:

https://github.com/leoge007/tdb-history-shadow-tools

It inventories large OpenClaw-style transcript archives, cleans them into strict seed inputs, runs the official seed flow into an isolated shadow directory, and audits L0/L1 counts, FTS/vector coverage, overlap, and live-memory pollution risk before any live-memory decision.

The goal is to make large historical memory backfills resumable, auditable, and safer to review before merging into a real memory store.

[leoge007]

补充一个中文说明：

我做了一个面向 TencentDB Agent Memory 的历史对话回填 shadow 工具链：先扫描大量 OpenClaw 历史 session/transcript，清洗成严格的 seed input，再把官方 seed 流程跑到隔离 shadow 目录，最后审计 L0/L1 数量、FTS/向量覆盖、重复/污染风险和 live memory 是否被误写。

它的定位不是替代 TDB，而是在大规模历史记忆导入前，加一层可批处理、可暂停、可复查的安全验证流程，避免直接把旧对话一次性灌进正式记忆库。

[leoge007]

Operational note / 使用前提：

This toolchain assumes your OpenClaw runtime already exposes the official TencentDB Agent Memory seed CLI. Before running a shadow seed, verify:

openclaw memory-tdai --help
openclaw memory-tdai seed --help

If these commands are unavailable, fix or upgrade the OpenClaw / plugin registration layer first. This repository intentionally does not patch OpenClaw, modify the TencentDB Agent Memory plugin, or bundle a private fallback runtime.

中文补充：正式使用前，需要先确认 TDB 的 memory-tdai CLI 已经正确注册到 OpenClaw。这个仓库只做历史数据清洗、shadow seed 和审计，不负责修改 OpenClaw 本体或修补插件注册层。

[Maxwell-Code07]

Hi @leoge007，这个历史对话回填工具做得挺好的 👍，在隔离环境先跑一遍再审计确认的思路很实用。感谢分享，欢迎继续共建！🎉

[leoge007]

插件的注册层，还是需要你们搞定啊 。不然就要自己改Openclaw的源码。

[Maxwell-Code07]

嗯嗯，已经看到对应的 issue #57 了，后续关于 CLI 注册的讨论我们在那边跟进哈 👌