nowledge-mem-openclaw-plugin/openclaw.plugin.json (1)

16-16: Consider mentioning the autoRecall dependency in the UI hint for maxRecallResults.

The configSchema description (line 46) correctly notes "only used when autoRecall is enabled", but the uiHints.help (line 16) omits this, which could confuse users who see the setting in the UI without understanding why changing it has no effect.

✏️ Suggested wording update

-			"help": "How many memories to inject for each recall cycle (1\u201320)"
+			"help": "How many memories to inject per turn when Auto-recall is on (1\u201320)"

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@nowledge-mem-openclaw-plugin/openclaw.plugin.json` at line 16, Update the
uiHints.help for the maxRecallResults setting to state that it only applies when
autoRecall is enabled; specifically, modify the "help" string (referenced by
uiHints.help for maxRecallResults) to include a note such as "Only used when
autoRecall is enabled" so it matches the configSchema description and prevents
user confusion.

nowledge-mem-openclaw-plugin/src/hooks/capture.js (1)

154-218: Inconsistent return sentinels from appendOrCreateThread.

Early-exit paths (no messages / empty normalized list) return undefined while error paths return null. The caller handles both correctly via !result, but the inconsistency makes the contract harder to reason about.

♻️ Suggested fix — unify all non-success paths to return null

-  if (!Array.isArray(rawMessages) || rawMessages.length === 0) return;
+  if (!Array.isArray(rawMessages) || rawMessages.length === 0) return null;
   ...
-  if (normalized.length === 0) return;
+  if (normalized.length === 0) return null;

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@nowledge-mem-openclaw-plugin/src/hooks/capture.js` around lines 154 - 218,
The function appendOrCreateThread has mixed non-success returns (undefined on
early exits and null on errors); make all non-success paths consistently return
null: change the early-exit returns after resolveHookMessages and after
filtering normalized to return null instead of undefined so every non-success
path (no messages, empty normalized, or failed append/create) returns null while
successful paths continue to return { threadId, normalized } and logs; update
the return statements in appendOrCreateThread accordingly.

nowledge-mem-openclaw-plugin/src/client.js (1)

697-723: title field in distillThread is not coerced to String.

All other fields in the same payload (thread_id, thread_content) use String(x || ""). thread_title: title || "" skips the String() coercion. While title is typed as string? in the JSDoc, the pattern is inconsistent.

♻️ Suggested fix

-            thread_title: title || "",
+            thread_title: String(title || ""),

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@nowledge-mem-openclaw-plugin/src/client.js` around lines 697 - 723, The
distillThread payload inconsistently sets thread_title using title || "" instead
of coercing to a string like the other fields; change the payload in
distillThread so thread_title is assigned String(title || "") to match thread_id
and thread_content (update the distillThread method where thread_title is set).

Configuration used: defaults

Review profile: CHILL

Plan: Pro

nowledge-mem-openclaw-plugin/src/hooks/capture.js (1)

154-217: Inconsistent implicit vs. explicit null returns in early-exit paths

Lines 156 and 163 return undefined (bare return;), while the error paths at lines 198 and 216 explicitly return null. The caller handles both correctly via !result, but normalising to a single sentinel would make the contract clearer.

♻️ Proposed fix

-	if (!Array.isArray(rawMessages) || rawMessages.length === 0) return;
+	if (!Array.isArray(rawMessages) || rawMessages.length === 0) return null;

 	// ...

-	if (normalized.length === 0) return;
+	if (normalized.length === 0) return null;

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@nowledge-mem-openclaw-plugin/src/hooks/capture.js` around lines 154 - 217,
The function appendOrCreateThread mixes implicit undefined returns (bare return)
and explicit null returns; change the early-exit returns after the rawMessages
check and after normalized.length === 0 to return null so all failure/early-exit
paths consistently return null (preserving existing success returns like {
threadId, normalized }); update any local comments or JSDoc for
appendOrCreateThread if present to reflect the unified null sentinel.

nowledge-mem-openclaw-plugin/src/tools/connections.js (2)

228-237: MENTIONS connections omit both node.id and strength — guideline compliance gap.

Every other edge type in this tool (CRYSTALLIZED_FROM, SOURCED_FROM, RELATED) shows the connected node's ID and/or strength. MENTIONS shows neither, even though both weight and node.id are available.

♻️ Proposed fix

- const lines = mentionConns.map(({ node }) => {
+ const lines = mentionConns.map(({ node, weight }) => {
      const name = getNodeTitle(node);
      const type = getNodeType(node);
-     return `  - ${name} (${type})`;
+     const strength = weight ? ` [${(weight * 100).toFixed(0)}%]` : "";
+     return `  - ${name} (${type})${strength}\n    → id: ${node.id}`;
  });

Based on learnings: "Tool nowledge_mem_connections must support graph exploration with edges: CRYSTALLIZED_FROM, EVOLVES (with sub-relations: supersedes/enriches/confirms/challenges), SOURCED_FROM, MENTIONS, showing strength % and memoryId."

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@nowledge-mem-openclaw-plugin/src/tools/connections.js` around lines 228 -
237, The MENTIONS branch omits node.id and weight; update the block handling
mentionConns (the grouped.get("MENTIONS") handling and the lines mapping that
uses getNodeTitle/node.getNodeType) to include the connected node's id and the
connection strength: for each ({ node, weight }) produce a line like "  - {name}
({type}) — id: {node.id}, strength: {Math.round(weight*100)}%" (or equivalent
percentage formatting) and push the section as before so MENTIONS matches the
other edge types by showing memoryId (node.id) and strength percent.

---

201-212: Two separate sections push the same "Knowledge evolution:" header — consider differentiating them.

If both graphExpand (lines 201–211) and graphEvolves (lines 282–303) return data, the output contains two adjacent "Knowledge evolution:" headings. Consider renaming one, e.g., "Knowledge evolution (neighbors):" vs "Knowledge evolution chain (full history):", or merging them.

Also applies to: 282-304

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@nowledge-mem-openclaw-plugin/src/tools/connections.js` around lines 201 -
212, The two blocks that push the same "Knowledge evolution:" header (the
graphExpand branch that builds evolvesConns and the graphEvolves branch) should
produce distinct headings so they don't appear duplicated; update the header
string used when pushing into sections in the evolvesConns block (identify by
variables grouped, evolvesConns, getNodeTitle, EVOLVES_RELATION_LABELS,
sections) to something like "Knowledge evolution (neighbors):" and keep or set
the other block (graphEvolves) to "Knowledge evolution chain (full history):" or
merge logic into a single section if preferred so the UI shows a single, clearly
labeled evolution area.

Configuration used: defaults

Review profile: CHILL

Plan: Pro

nowledge-mem-openclaw-plugin/src/hooks/capture.js (1)

224-228: buildConversationText still unbounded — previous review concern unresolved

buildConversationText maps over fullContent (untruncated raw text) with no ceiling. The only guard downstream is a 100-char floor at line 272. For sessions with large code blocks, the payload sent to both triageConversation and distillThread can still far exceed typical LLM context or HTTP payload limits.

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@nowledge-mem-openclaw-plugin/src/hooks/capture.js` around lines 224 - 228,
buildConversationText currently concatenates m.fullContent with no size cap,
letting very large messages (e.g., long code blocks) blow past LLM/HTTP limits;
change it to use a bounded/truncated variant of each message before joining
(e.g., prefer an existing m.truncatedContent or apply a truncate helper to
m.fullContent with a safe per-message max like 2k chars), so the outputs passed
to triageConversation and distillThread are size-limited; update
buildConversationText to reference the truncation (and ensure the helper
preserves useful context such as head/tail of code) rather than using
m.fullContent directly.

nowledge-mem-openclaw-plugin/src/hooks/capture.js (1)

257-260: Optional: add a null guard on cfg for direct-call safety

buildAgentEndCaptureHandler is exported. If it is ever invoked directly (e.g. in tests) without a config object, cfg.autoCapture at line 260 throws a TypeError rather than cleanly skipping distillation.

🛡️ Proposed fix

 export function buildAgentEndCaptureHandler(client, cfg, logger) {
+	const _cfg = cfg ?? {};
 	return async (event, ctx) => {
 		if (!event?.success) return;

 		// 1. Always thread-append (idempotent, self-guards on empty messages).
 		const result = await appendOrCreateThread({
 			client,
 			logger,
 			event,
 			ctx,
 			reason: "agent_end",
 		});

 		// 2. Triage + distill
-		if (!cfg.autoCapture) return;
+		if (!_cfg.autoCapture) return;

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@nowledge-mem-openclaw-plugin/src/hooks/capture.js` around lines 257 - 260,
The handler checks cfg.autoCapture without guarding cfg, which can throw if
buildAgentEndCaptureHandler is called with no config; add a null/undefined guard
(e.g., if (!cfg || !cfg.autoCapture) return) at the start of the distillation
block so the function safely returns when cfg is falsy; update the check around
cfg.autoCapture in buildAgentEndCaptureHandler to handle a missing cfg object.

nowledge-mem-openclaw-plugin/src/client.js (1)

82-84: Hardcoded default URL repeated three times — extract a local constant

"http://127.0.0.1:14242" appears verbatim at lines 83, 172, and 186. config.js already exports API_DEFAULT_URL for this exact value, but client.js doesn't import it. At minimum, define a module-local constant to avoid the triple repetition.

♻️ Proposed fix

 import { spawnSync } from "node:child_process";

+const DEFAULT_API_URL = "http://127.0.0.1:14242";
+
 // ...

-		this._apiUrl =
-			(credentials.apiUrl || "").trim() || "http://127.0.0.1:14242";
+		this._apiUrl =
+			(credentials.apiUrl || "").trim() || DEFAULT_API_URL;

 // ...in _spawnEnv:
-		if (this._apiUrl !== "http://127.0.0.1:14242") {
+		if (this._apiUrl !== DEFAULT_API_URL) {

 // ...in _apiUrlArgs:
-		return this._apiUrl !== "http://127.0.0.1:14242"
+		return this._apiUrl !== DEFAULT_API_URL

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@nowledge-mem-openclaw-plugin/src/client.js` around lines 82 - 84, The
hardcoded default URL "http://127.0.0.1:14242" is repeated in client.js (used to
set this._apiUrl) and should be centralized: import or reference the existing
API_DEFAULT_URL from config.js (or define a module-local constant named
API_DEFAULT_URL) and replace the literal occurrences where this._apiUrl is
assigned (currently using (credentials.apiUrl || "").trim() ||
"http://127.0.0.1:14242") and any other places that repeat the URL (lines around
172 and 186) to use the imported/constant value instead; ensure trimming and
fallback logic remains the same (e.g., (credentials.apiUrl || "").trim() ||
API_DEFAULT_URL) and update any related tests or usages if necessary.

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Configuration used: defaults

Review profile: CHILL

Plan: Pro

nowledge-mem-openclaw-plugin/src/hooks/capture.js (1)

156-220: Inconsistent early-return types: undefined vs null.

Lines 158 and 165 return undefined implicitly (no messages / no valid normalized messages), while lines 200 and 218 return null explicitly on errors. The downstream guard at line 278 (!result) handles both correctly, so this isn't a bug. However, a consistent return convention (e.g., always null for "no result") would improve clarity for future callers.

♻️ Suggested: make early returns explicit

 async function appendOrCreateThread({ client, logger, event, ctx, reason }) {
 	const rawMessages = await resolveHookMessages(event);
-	if (!Array.isArray(rawMessages) || rawMessages.length === 0) return;
+	if (!Array.isArray(rawMessages) || rawMessages.length === 0) return null;
 
 	const threadId = buildStableThreadId(event, ctx);
 	const sessionKey = String(ctx?.sessionKey || ctx?.sessionId || "session");
 	const sessionId = String(ctx?.sessionId || "").trim();
 	const title = buildThreadTitle(ctx, reason);
 	const normalized = rawMessages.map(normalizeRoleMessage).filter(Boolean);
-	if (normalized.length === 0) return;
+	if (normalized.length === 0) return null;

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@nowledge-mem-openclaw-plugin/src/hooks/capture.js` around lines 156 - 220,
The function appendOrCreateThread has inconsistent early-return types (implicit
undefined vs explicit null); change the early exits that currently do "return"
(after resolving rawMessages and after filtering normalized) to explicitly
"return null" so all non-success paths return null, keeping the existing null
returns in the catch blocks unchanged; update the two locations that check
rawMessages and normalized within appendOrCreateThread (the branches using
resolveHookMessages and normalized.map/filter) to return null.

Configuration used: defaults

Review profile: CHILL

Plan: Pro