fix: update KnowledgeService call sites and add knowledge cleanup migration

Pass embedding service to KnowledgeService::new() at all call sites (mcp,
serve, degraded_mode test). Align CLI auto_archive retention to 90 days.
Add migration to strip UUIDs from knowledge titles, merge duplicates, and
fix null/invalid project names in observations.

Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-opencode)

Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>

Author: Stranmor

crates/cli/src/commands/mcp.rs

@@ -73,7 +73,7 @@ pub(crate) async fn run(config: Arc<AppConfig>) -> Result<()> {
         &config,
     ));
     let session_service = Arc::new(SessionService::new(storage.clone(), llm.clone()));
-    let knowledge_service = Arc::new(KnowledgeService::new(storage.clone()));
+    let knowledge_service = Arc::new(KnowledgeService::new(storage.clone(), embeddings.clone()));
     let search_service = Arc::new(SearchService::new(
         storage,
         embeddings,

crates/cli/src/commands/search.rs

@@ -124,7 +124,7 @@ pub(crate) async fn run_backfill_embeddings(batch_size: usize) -> Result<()> {
 pub(crate) async fn run_knowledge_lifecycle() -> Result<()> {
     let storage = crate::create_storage_from_env().await?;
     let decayed = storage.decay_confidence().await?;
-    let archived = storage.auto_archive(30).await?;
+    let archived = storage.auto_archive(90).await?;
     println!("Knowledge confidence lifecycle complete:");
     println!("  Entries with decayed confidence: {decayed}");
     println!("  Entries archived: {archived}");

crates/cli/src/commands/serve.rs

@@ -80,7 +80,7 @@ pub(crate) async fn run(port: u16, host: String, config: Arc<AppConfig>) -> Resu
         &config,
     ));
     let session_service = Arc::new(SessionService::new(storage.clone(), llm.clone()));
-    let knowledge_service = Arc::new(KnowledgeService::new(storage.clone()));
+    let knowledge_service = Arc::new(KnowledgeService::new(storage.clone(), embeddings.clone()));
     let search_service = Arc::new(SearchService::new(
         storage.clone(),
         embeddings.clone(),

crates/mcp/tests/degraded_mode.rs

@@ -82,7 +82,7 @@ fn setup_degraded_services() -> (
         &config,
     ));
     let session_service = Arc::new(SessionService::new(backend.clone(), llm.clone()));
-    let knowledge_service = Arc::new(KnowledgeService::new(backend.clone()));
+    let knowledge_service = Arc::new(KnowledgeService::new(backend.clone(), None));
     let search_service = Arc::new(SearchService::new(
         backend,
         None,

crates/storage/migrations/20260315000001_knowledge_cleanup.sql

@@ -0,0 +1,76 @@
+-- Knowledge cleanup migration: strip UUIDs from titles, merge duplicates,
+-- fix null/invalid project names.
+
+-- 1. Strip UUID suffixes from knowledge titles.
+-- Pattern: optional whitespace + hex UUID (8-4-4+ with optional trailing segments).
+UPDATE global_knowledge
+SET title = TRIM(REGEXP_REPLACE(title, '\s*[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4,}(-[0-9a-f]*)*\s*', '', 'gi')),
+    updated_at = NOW()
+WHERE title ~ '[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4,}';
+
+-- 2. Merge duplicate knowledge entries (same lowercased title after UUID stripping).
+-- Keep the entry with highest usage_count (ties broken by earliest created_at).
+-- Sum usage_counts, merge source_observations arrays, take highest confidence.
+DO $$
+DECLARE
+    dup RECORD;
+    keeper_id TEXT;
+    merged_usage BIGINT;
+    merged_obs JSONB;
+    max_confidence FLOAT8;
+BEGIN
+    -- Find groups of duplicates (2+ entries with same normalized title)
+    FOR dup IN
+        SELECT LOWER(TRIM(title)) AS norm_title,
+               COUNT(*) AS cnt
+        FROM global_knowledge
+        WHERE archived_at IS NULL
+        GROUP BY LOWER(TRIM(title))
+        HAVING COUNT(*) > 1
+    LOOP
+        -- Determine the keeper: highest usage_count, then earliest created_at
+        SELECT id INTO keeper_id
+        FROM global_knowledge
+        WHERE LOWER(TRIM(title)) = dup.norm_title AND archived_at IS NULL
+        ORDER BY usage_count DESC, created_at ASC
+        LIMIT 1;
+
+        -- Aggregate usage_count, source_observations, and max confidence from all duplicates
+        SELECT COALESCE(SUM(usage_count), 0),
+               COALESCE(
+                   (SELECT jsonb_agg(DISTINCT elem)
+                    FROM global_knowledge g2,
+                         jsonb_array_elements(g2.source_observations) AS elem
+                    WHERE LOWER(TRIM(g2.title)) = dup.norm_title
+                      AND g2.archived_at IS NULL
+                      AND elem != 'null'::jsonb),
+                   '[]'::jsonb
+               ),
+               COALESCE(MAX(confidence), 0.5)
+        INTO merged_usage, merged_obs, max_confidence
+        FROM global_knowledge
+        WHERE LOWER(TRIM(title)) = dup.norm_title AND archived_at IS NULL;
+
+        UPDATE global_knowledge
+        SET usage_count = merged_usage,
+            source_observations = merged_obs,
+            confidence = max_confidence,
+            updated_at = NOW()
+        WHERE id = keeper_id;
+
+        DELETE FROM global_knowledge
+        WHERE LOWER(TRIM(title)) = dup.norm_title
+          AND archived_at IS NULL
+          AND id != keeper_id;
+    END LOOP;
+END $$;
+
+-- 3. Fix NULL project in observations — set to 'unknown'.
+UPDATE observations
+SET project = 'unknown'
+WHERE project IS NULL;
+
+-- 4. Fix 'ivan plankin' project name (personal name, not a project).
+UPDATE observations
+SET project = 'unknown'
+WHERE LOWER(project) = 'ivan plankin';