fix(openclaw-plugin): default recall to assemble and bound afterTurn

examples/openclaw-plugin/INSTALL.md

@@ -92,6 +92,17 @@ Get the current full plugin configuration:
 openclaw config get plugins.entries.openviking.config
 ```
 
+Recommended default for the context-engine path:
+
+```bash
+openclaw config get plugins.entries.openviking.config.recallPath
+openclaw config get plugins.entries.openviking.hooks.allowPromptInjection
+```
+
+The supported default is `recallPath = assemble` with
+`hooks.allowPromptInjection = false`, so memory recall stays inside
+`assemble()` instead of running in `before_prompt_build`.
+
 ### Local Mode
 
 Use this mode when the OpenClaw plugin should start and manage a local OpenViking process.

examples/openclaw-plugin/README.md

@@ -51,18 +51,18 @@ The main rules are:
 
 This matters because the plugin is built to support multi-agent and multi-session OpenClaw usage without mixing memories across sessions.
 
-## Prompt-Front Recall Flow
+## Recall Flow
 
 ![Automatic recall flow before prompt build](./images/openclaw-plugin-recall-flow.png)
 
-Today the main recall path still lives in `before_prompt_build`:
+The default recall path now lives in `assemble()`:
 
-1. Extract the latest user text from `messages` or `prompt`.
+1. Extract the latest user text from the active messages passed into `assemble()`.
 2. Resolve the agent routing for the current `sessionId/sessionKey`.
-3. Run a quick availability precheck so prompt building does not stall when OpenViking is unavailable.
+3. Read session context from OpenViking under the configured token budget.
 4. Query both `viking://user/memories` and `viking://agent/memories` in parallel.
 5. Deduplicate, threshold-filter, rerank, and trim the results under a token budget.
-6. Prepend the selected memories as a `<relevant-memories>` block.
+6. Append the selected memories into `systemPromptAddition` as a `<relevant-memories>` block.
 
 The reranking logic is not pure vector-score sorting. The current implementation also considers:
 
@@ -79,7 +79,7 @@ When the latest user input looks like pasted multi-speaker transcript content:
 
 - metadata blocks, command text, and pure question text are filtered out
 - the cleaned text is checked against speaker-turn and length thresholds
-- if it matches, the plugin prepends a lightweight `<ingest-reply-assist>` instruction
+- if it matches, the plugin adds a lightweight `<ingest-reply-assist>` instruction to `systemPromptAddition`
 
 The goal is not to change memory logic. It is to reduce the chance that the model responds with `NO_REPLY` when the user pastes chat history, meeting notes, or conversation transcripts for ingestion.
 
@@ -100,7 +100,7 @@ Session handling is the main axis of this design. In the current implementation
 - tool output becomes separate `toolResult`
 - the final message list goes through a tool-use/result pairing repair pass
 
-That means OpenClaw sees “compressed history summary + archive index + active messages”, not an ever-growing raw transcript.
+That means OpenClaw sees “compressed history summary + archive index + active messages”, not an ever-growing raw transcript. When recall is enabled, `assemble()` also becomes the main memory-injection surface for fresh-session questions.
 
 ### What `afterTurn()` does
 
@@ -205,8 +205,8 @@ The main difference between `local` and `remote` is who is responsible for bring
 The repo also contains a more future-looking design draft at `docs/design/openclaw-context-engine-refactor.md`. It is important not to conflate the two:
 
 - this README describes current implemented behavior
-- the older draft discusses a stronger future move into context-engine-owned lifecycle control
-- in the current version, the main automatic recall path still lives in `before_prompt_build`, not fully in `assemble()`
+- the older draft discusses a broader context-engine-owned lifecycle control plan
+- in the current version, automatic recall defaults to `assemble()` and `before_prompt_build` is a compatibility path only
 - in the current version, `afterTurn()` already appends to the OpenViking session, but commit remains threshold-triggered and asynchronous on that path
 - in the current version, `compact()` already uses `commit(wait=true)`, but it is still focused on synchronous commit plus readback rather than owning every orchestration concern
 

examples/openclaw-plugin/config.ts

@@ -18,6 +18,7 @@ export type MemoryOpenVikingConfig = {
   captureMode?: "semantic" | "keyword";
   captureMaxLength?: number;
   autoRecall?: boolean;
+  recallPath?: "assemble" | "hook";
   recallLimit?: number;
   recallScoreThreshold?: number;
   recallMaxContentChars?: number;
@@ -46,6 +47,7 @@ const DEFAULT_TIMEOUT_MS = 15000;
 const DEFAULT_CAPTURE_MODE = "semantic";
 const DEFAULT_CAPTURE_MAX_LENGTH = 24000;
 const DEFAULT_RECALL_LIMIT = 6;
+const DEFAULT_RECALL_PATH = "assemble";
 const DEFAULT_RECALL_SCORE_THRESHOLD = 0.15;
 const DEFAULT_RECALL_MAX_CONTENT_CHARS = 500;
 const DEFAULT_RECALL_PREFER_ABSTRACT = true;
@@ -154,6 +156,7 @@ export const memoryOpenVikingConfigSchema = {
         "captureMode",
         "captureMaxLength",
         "autoRecall",
+        "recallPath",
         "recallLimit",
         "recallScoreThreshold",
         "recallMaxContentChars",
@@ -196,6 +199,14 @@ export const memoryOpenVikingConfigSchema = {
     ) {
       throw new Error(`openviking captureMode must be "semantic" or "keyword"`);
     }
+    const recallPath = cfg.recallPath;
+    if (
+      typeof recallPath !== "undefined" &&
+      recallPath !== "assemble" &&
+      recallPath !== "hook"
+    ) {
+      throw new Error(`openviking recallPath must be "assemble" or "hook"`);
+    }
 
     return {
       mode,
@@ -213,6 +224,7 @@ export const memoryOpenVikingConfigSchema = {
         Math.min(200_000, Math.floor(toNumber(cfg.captureMaxLength, DEFAULT_CAPTURE_MAX_LENGTH))),
       ),
       autoRecall: cfg.autoRecall !== false,
+      recallPath: recallPath ?? DEFAULT_RECALL_PATH,
       recallLimit: Math.max(1, Math.floor(toNumber(cfg.recallLimit, DEFAULT_RECALL_LIMIT))),
       recallScoreThreshold: Math.min(
         1,
@@ -334,6 +346,12 @@ export const memoryOpenVikingConfigSchema = {
       label: "Auto-Recall",
       help: "Inject relevant OpenViking memories into agent context",
     },
+    recallPath: {
+      label: "Recall Path",
+      placeholder: DEFAULT_RECALL_PATH,
+      advanced: true,
+      help: '"assemble" keeps memory injection inside the context-engine path; "hook" preserves legacy before_prompt_build recall.',
+    },
     recallLimit: {
       label: "Recall Limit",
       placeholder: String(DEFAULT_RECALL_LIMIT),