feat(agent-memory): Phase 8 — self-optimization config reads

[jamby77]

Stacked on #255 (Phase 7 — discovery marker).

What

Phase 8 of @betterdb/agent-memory: runtime-tunable recall/eviction knobs, so BetterDB Monitor's cache-proposals engine can retune a memory store without a restart (same pattern as semantic-cache's configRefresh).

MemoryStore reads {name}:__mem_config and live-applies:

• recall.threshold (cosine-distance ceiling, 0..2)

• recall.weights.similarity / recall.weights.recency / recall.weights.importance

• recall.halfLifeSeconds

• maxItemsPerScope

• refreshConfig() (public, manual tick) reads the hash and applies it; currentConfig() exposes the effective tunables.

• Opt-in configRefresh?: boolean | { enabled?, intervalMs? } enables an immediate read plus an unref'd interval (default 30s, min 1s); close() stops it.

Design / review notes

• Off by default. Unlike semantic-cache (default-on), refresh is opt-in here so a standalone store never polls and existing unit tests keep their exact .call sequences.
• Snapshot fallback. Each refresh rebuilds the config from the constructor values plus only the fields present in the hash, so removing a field reverts to the constructor default.
• Validation. Invalid/out-of-range/NaN values are ignored; reads are best-effort and never throw. An all-zero weight vector is rejected (it would make every composite score 0 and leave recall ordering undefined) — kept from a review finding. initialWeights is copied so the shared DEFAULT_WEIGHTS constant can't be aliased.
• recall() captures threshold+weights into locals before its first await, so an interval refresh can't tear an in-flight recall.

Tests (MemoryStore.config.test.ts, 12)

defaults snapshot · threshold · weights · halfLifeSeconds+maxItemsPerScope · partial config leaves others default · field-removal reverts · invalid ignored · all-zero weights rejected · live threshold affects recall end-to-end · no polling when disabled · immediate+interval read & stop on close (fake timers) · best-effort on read failure.

77/77 package tests green · tsc clean · prettier clean.

---

Note

Medium Risk
Changes live recall scoring, eviction victim selection, and capacity limits when external config is applied; validation and snapshots limit bad states, but mis-tuned remote values can still alter memory behavior at runtime.

Overview
Adds opt-in runtime tuning for MemoryStore by reading {name}:__mem_config from Valkey, mirroring semantic-cache’s configRefresh pattern so Monitor can retune recall/eviction without a restart.

refreshConfig() and currentConfig() expose manual reads and the effective snapshot (threshold, weights, half-life, maxItemsPerScope). With configRefresh enabled, the store polls on an interval (default 30s, min 1s), runs an immediate read at startup, and close() stops the timer. Refresh is off by default so existing stores never poll unless opted in.

Each refresh rebuilds from constructor defaults plus only fields present in the hash; removed fields revert. Invalid values are ignored; all-zero weight vectors are rejected. recall() and enforceCapacity() snapshot tunables before async work so concurrent refresh cannot mix config versions mid-operation.

New MemoryStore.config.test.ts coverage and an eviction test for mid-pass refresh; public types exported from index.ts.

^{Reviewed by Cursor Bugbot for commit 7d0fa87. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 37c7a4b. Configure here.}