M3: Dataset snapshot + tags (versioning safety net) for openclaw-mem-engine

[phenomenoner]

Summary

We have strong system-level rollback today (switch plugins.slots.memory), but we lack a dataset-level versioning safety net for the openclaw-mem-engine LanceDB dataset.

This issue proposes an M3 v0 implementation: filesystem snapshots + manifest with operator-grade commands to create/list/checkout/delete snapshots (tagged).

Spec

• Draft spec committed: docs/specs/mem-engine-versioning-safety-net-v0.md
• Commit: 7a06818

Goals (v0)

• Snapshot LanceDB dbPath before risky behavior/data changes (auto-capture rule changes, regrading, mass writeback, reindex/optimize)
• Deterministic, bounded receipts (no raw memory text)
• Safe defaults: tag validation + path traversal prevention + fail-closed for destructive ops

Acceptance checklist

• Create snapshot + manifest + receipt
• List snapshots deterministically
• Checkout switches engine dbPath to a snapshot and remains functional
• One-screen rollback instructions
• Tests: tag validation + traversal prevention

Notes

Implementation preference: python CLI in openclaw-mem for filesystem ops; engine config update via openclaw config set (optional) + gateway restart as explicit step.}

[phenomenoner]

Closing as implemented.

Receipts:

• Engine snapshot CLI supports create/list/checkout/delete with manifests.
• Tests cover roundtrip, symlink skipping, checkout backup restore after partial copy failure, and path traversal tag rejection.
• Latest verification: tests/test_engine_snapshots.py passed as part of the targeted suite before push.

Related current verification included 28 targeted tests passing before push.