Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/main/java/ai/labs/eddi/configs/agents/model/AgentConfiguration.java`:
- Around line 529-560: The new summarization fields (summarizeMinEntries,
summarizeTargetEntries, summarizeGroupBy, preserveAgentProvenance,
maxSummarizationCalls, summarizationPrompt) should be extracted into an
immutable nested record to follow the project's config guideline; create a
record SummarizationConfig(int minEntries,int targetEntries,String
groupBy,boolean preserveAgentProvenance,int maxCalls,String prompt) with a
no-arg canonical default constructor that sets the current defaults
(5,2,"category",false,10 and the existing prompt), replace the six mutable
fields in DreamConfig with a single private SummarizationConfig summarization =
new SummarizationConfig(), add a public getter (and a setter on DreamConfig that
replaces the whole SummarizationConfig if project pattern requires mutability at
the container level), remove the original six fields, and update any code that
referenced summarizeMinEntries, summarizeTargetEntries, summarizeGroupBy,
preserveAgentProvenance, maxSummarizationCalls or summarizationPrompt to read
from summarization.minEntries(), summarization.targetEntries(), etc.

---

Nitpick comments:
In `@src/main/java/ai/labs/eddi/engine/runtime/internal/DreamService.java`:
- Line 60: The DreamService currently instantiates a new ObjectMapper via the
field "private final ObjectMapper objectMapper = new ObjectMapper();"; replace
this with CDI injection of the Quarkus-managed ObjectMapper (e.g. remove the
"new ObjectMapper()" instantiation and add an injected ObjectMapper either as a
constructor parameter or a field annotated for injection) so DreamService uses
the application-configured Jackson instance; update constructor and usages to
reference the injected "objectMapper" field (keep the same symbol name to
minimize other changes).

                for (var entry : consolidated) {
                    userMemoryStore.upsert(new UserMemoryEntry(
                            null, userId, entry.key(), entry.value(),
                            groupEntries.getFirst().category(),

                Visibility mergedVisibility = mostRestrictiveVisibility(groupEntries);
                String sourceAgent = groupEntries.getFirst().sourceAgentId();
                Instant earliestCreated = groupEntries.stream()

                    userMemoryStore.upsert(new UserMemoryEntry(
                            null, userId, entry.key(), entry.value(),
                            groupEntries.getFirst().category(),
                            mergedVisibility, sourceAgent, List.of(),
                            "dream-consolidation", false, 0,

            // 6. Cap at target
            if (consolidated.size() > config.getSummarizeTargetEntries()) {
                consolidated = consolidated.subList(0, config.getSummarizeTargetEntries());
            }

                for (var entry : consolidated) {
                    userMemoryStore.upsert(new UserMemoryEntry(
                            null, userId, entry.key(), entry.value(),
                            groupEntries.getFirst().category(),
                            mergedVisibility, sourceAgent, List.of(),
                            "dream-consolidation", false, 0,
                            earliestCreated, Instant.now()));
                }
            } catch (Exception e) {
                LOGGER.warnf("[DREAM] Failed to insert consolidated entries for " +
                        "user='%s', group='%s': %s. Originals preserved.",
                        userId, group.getKey(), e.getMessage());
                continue; // Insert failed → don't delete anything

            for (var original : groupEntries) {
                try {
                    userMemoryStore.deleteEntry(original.id());
                } catch (Exception e) {
                    LOGGER.warnf("[DREAM] Failed to delete original entry '%s': %s. " +
                            "Duplicate will be resolved by contradiction detector.",
                            original.id(), e.getMessage());
                }
            }

            int reduced = groupEntries.size() - consolidated.size();
            totalConsolidated += reduced;
            entriesSummarizedCounter.increment(reduced);

                for (var entry : consolidated) {
                    userMemoryStore.upsert(new UserMemoryEntry(
                            null, userId, entry.key(), entry.value(),
                            groupEntries.getFirst().category(),
                            mergedVisibility, sourceAgent, List.of(),

        if ("all".equals(config.getSummarizeGroupBy())) {
            // Single group
            return Map.of("all", new ArrayList<>(entries));
        }

                    LOGGER.warnf("[DREAM] Failed to delete original entry '%s': %s. " +
                            "Duplicate will be resolved by contradiction detector.",

            return entries.stream()
                    .filter(m -> m.containsKey("key") && m.containsKey("value"))
                    .map(m -> new ConsolidatedEntry(m.get("key"), m.get("value")))

                for (var entry : consolidated) {
                    userMemoryStore.upsert(new UserMemoryEntry(
                            null, userId, entry.key(), entry.value(),

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/main/java/ai/labs/eddi/modules/llm/impl/SummarizationService.java`:
- Around line 148-149: The catch block in SummarizationService that currently
calls LOGGER.warnf to record summarization failures should use LOGGER.errorf
instead; locate the logging call in SummarizationService (the LOGGER.warnf(e,
"[SUMMARIZATION] Failed to summarize: provider=%s, model=%s, error=%s",
llmProvider, llmModel, e.getMessage()) invocation) and replace warnf with errorf
so failures are logged at ERROR level while keeping the same message,
parameters, and exception argument.

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@src/main/java/ai/labs/eddi/modules/llm/impl/SummarizationService.java`:
- Around line 151-155: In SummarizationService's RuntimeException catch block
(the handler that calls errorCounter.increment() and currently uses
LOGGER.warnf), change the logging call from LOGGER.warnf(...) to
LOGGER.errorf(...) so runtime failures are logged at ERROR level with the same
formatted message (include llmProvider, llmModel, and e.getMessage()), and keep
the increment and rethrow behavior intact.