fix(openclaw): migrate example plugin memory search to the v1 POST + Filters DSL API

[Fearvox]

What

The OpenClaw example plugin's memory recall never worked against the v1
backend. searchMemories (examples/openclaw-plugin/src/api.js) still spoke the
v0 dialect on both the request and the response side:

Request — it sent GET /api/v1/memories/search with a flat top-level
user_id and a retrieve_method field. Against the v1 API that fails twice:

• the route is POST-only, so the GET returns 405 Method Not Allowed;
• even as a POST, SearchMemoriesRequest requires a Filters DSL object
  (filters must carry user_id/group_id at the first level), so a flat
  top-level user_id returns 422 "Field required: filters";
• the retrieval-method field is method, not retrieve_method.

Response — it read r.result.memories / r.result.pending_messages, but the
v1 response is { data: { episodes, profiles, raw_messages, ... } }. So even a
successful call surfaced zero memories.

This PR migrates searchMemories to the v1 contract end-to-end while preserving
the function's return shape (so parseSearchResponse and the assembler are
untouched):

• Build the v1 request envelope: POST + filters: { user_id, group_id? } +
  method (mapped from retrieve_method) + query / top_k / memory_types.
• Map the v1 response: data.episodes → memories (tagging memory_type: "episodic_memory" so the downstream episodic filter matches; carrying
  summary/episode/subject/score/timestamp), and data.raw_messages → pending_messages (flattening content_items to text).

Verification

Live against a running EverCore (:1995):

• old form GET …/search → 405; flat top-level user_id (POST) → 422
  "Field required: filters".
• the fixed body (POST + filters:{user_id} + method) → HTTP 200, with
  the response echoing "filters_applied": {"user_id": "…"}.
• the real branch searchMemories() driven end-to-end against :1995
  (method=keyword) returns status: "ok" and maps the response with no error.

Note: method=hybrid/vector additionally require a reachable embedding
backend; on a host without one they 500 at the embedding step (an environment
dependency, not this code). keyword (ES-only) needs no embeddings and was
used for the live 200.

Offline regression — adds test/search-memories.test.js (node --test, no
live stack):

• asserts the request is a POST whose body carries filters: { user_id } (no
  top-level user_id) and method (no retrieve_method), with only
  backend-searchable memory_types;
• feeds a canned v1 { data: { episodes, raw_messages } } response and asserts
  it maps into { memories, pending_messages } and that the existing
  parseSearchResponse consumes the mapped shape into the expected episodic /
  pending text.

cd methods/EverCore/examples/openclaw-plugin
node --test test/search-memories.test.js

Result: 2 passed.

Scope

Surgical: src/api.js (searchMemories only) + its new regression test. The v1
demo migration (#191, PR #196) covered methods/evermemos/demo/* and docs but
not this example plugin, so the plugin's search path was left on v0. This PR
finishes that migration for the plugin. Does not touch saveMemories (shipped
separately as the #237 fix).

Credit

Builds on prior art:

• CZH-THU (docs: correct openclaw plugin api reference endpoints #202) — documented the correct OpenClaw plugin API reference
  endpoints (the POST /search route this aligns the client to).

Co-authored-by: CZH-THU CZH-THU@users.noreply.github.com
Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com

🤖 Generated with Claude Code

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.