[integrations] Add OpenClaw Agent Memory tool schemas

[Humestone]

Summary

• replace broad Type.Record(Type.String(), Type.Any()) parameters for openbrain_recall and openbrain_writeback with explicit Typebox schemas
• mirror the Agent Memory API request shapes for recall scope/limits/entities and write-back memory payload/provenance/source references
• rebuild integrations/openclaw-agent-memory/plugin/dist/index.js from the updated source

Why

This keeps the OpenClaw plugin tool surface inspectable and safer for agents without changing the API client behavior. It is intentionally complementary to #278: this PR only describes the tool parameters; #278 can still inject schema versions and improve API error detail.

Verification

• npm install --ignore-scripts --omit=peer
• npm run build
• node --check dist/index.js
• git diff --check

Notes

• workspace_id and schema_version remain optional at the tool boundary because the plugin/client path and pending [integrations] OpenClaw plugin: inject schema_version + surface error details in client #278 own runtime injection behavior.
• This does not add new tools, change network behavior, or touch live Agent Memory data.

Related to #278, #279.

[alanshurafa]

Thanks for the contribution. The explicit Typebox schemas for openbrain_recall and openbrain_writeback mirror the Edge Function's expected shapes closely, and keeping schema_version optional at the tool boundary is the right call — runtime injection is the client's job. The bounded limits (max_items and max_tokens ranges) are a good touch for constraining the model. dist/index.js is regenerated, so the built plugin reflects the change.

One coordination note for the maintainer: #309 modifies src/index.ts, which #281 and #283 also touch — whichever of the cluster lands first, the others will need a small rebase. Merge ordering across the four-PR OpenClaw cluster (#278, #281, #283, #309) is flagged separately. Recommend maintainer review.

— Alan (community reviewer; non-binding)