akm-cli 0.9.17-alpha.4 → 0.9.17-alpha.6

package/dist/storage/repositories/index-entries-repository.js

@@ -174,12 +174,11 @@ export function getBaseBeliefStatesForDerivedTwins(db, twinIds) {
 /**
  * Re-key an entries row in place for the opt-in source-maintenance script.
  *
- * The row id is preserved on purpose — `utility_scores`,
- * `utility_scores_scoped`, and `embeddings` are keyed by `entry_id`, so an
- * UPDATE (rather than a delete + insert under the new `item_ref`) is what
- * keeps the asset's accumulated usage-ranking history attached across a
- * rename. (`asset_salience` / `asset_outcome` live in state.db keyed by
- * `asset_ref` TEXT and are re-keyed separately by `akm mv` — see
+ * The row id is preserved on purpose — `utility_scores` and `embeddings` are
+ * keyed by `entry_id`, so an UPDATE (rather than a delete + insert under the
+ * new `item_ref`) is what keeps the asset's accumulated usage-ranking history
+ * attached across a rename. (`asset_salience` / `asset_outcome` live in
+ * state.db keyed by `asset_ref` TEXT and are re-keyed separately by `akm mv` — see
  * the state rekey helper.) `document_json.name` (and `filename`, when
  * present) is patched and `embed_hash` recomputed so search reflects the new
  * name. Its FTS projection and stale vector are updated in the same
@@ -407,7 +406,6 @@ function deleteRelatedRows(db, ids, options = {}) {
         bestEffort(() => db.prepare(`DELETE FROM embeddings WHERE id IN (${placeholders})`).run(...chunk), "delete embeddings for entries");
         // Clean up utility scores before deleting entries
         bestEffort(() => db.prepare(`DELETE FROM utility_scores WHERE entry_id IN (${placeholders})`).run(...chunk), "delete utility_scores for entries");
-        bestEffort(() => db.prepare(`DELETE FROM utility_scores_scoped WHERE entry_id IN (${placeholders})`).run(...chunk), "delete utility_scores_scoped for entries");
     }
     // usage_events lives in state.db, outside this transaction. Index persistence
     // disables this cleanup and runs it only after its index.db transaction

package/dist/storage/repositories/index-schema.js

@@ -10,8 +10,9 @@
  * for columns added after a table first shipped, drops of retired derived
  * tables and columns, and one in-place rebuild of the (derived, cheap) FTS
  * table when its layout is older than this release's. It never drops
- * `entries`, `embeddings`, `utility_scores*`, `graph_*`, or
- * `llm_enrichment_cache` to cross a version boundary; the only from-scratch
+ * `entries`, `embeddings`, `utility_scores`, the extracted graph
+ * (`graph_meta`, `graph_files`, `graph_file_*`), or `llm_enrichment_cache`
+ * to cross a version boundary; the only from-scratch
  * rebuild is the SQLITE_CORRUPT path in `index-connection.ts`. A layout newer
  * than this release's is refused, naming the upgrade
  * ({@link newerIndexLayoutError}).
@@ -27,7 +28,11 @@ import { getMeta, setMeta } from "./index-meta-repository.js";
 export const DB_VERSION = CANONICAL_INDEX_DB_VERSION;
 /** `index_meta` key set when the writable opener migrated the layout; cleared once `akm index` VACUUMs. */
 export const VACUUM_PENDING_META = "vacuumPending";
-// #624-P1: graph_files is keyed to (stash_root, file_path, body_hash).
+/**
+ * The value written to `graph_meta.schema_version`, a NOT NULL column in every
+ * released layout. Releases up to 0.9.17-alpha.5 write 4 and nothing ever
+ * compares it; the index layout version gates the graph tables' shape.
+ */
 export const GRAPH_SCHEMA_VERSION = 4;
 /**
  * The refusal for an index a newer akm wrote. Readers and the writable opener
@@ -62,7 +67,7 @@ const REGISTRY_INDEX_CACHE_DDL = `
 `;
 /**
  * Create the graph-extraction tables (`graph_meta`/`graph_files`/`graph_file_entities`/
- * `graph_file_relations`/`graph_extraction_queue`).
+ * `graph_file_relations`).
  *
  * graph_files is self-keyed on (stash_root, file_path, body_hash) and is not
  * tied to entries.id (#624-P1): re-upserting an entries row never disturbs the
@@ -141,21 +146,6 @@ function ensureGraphTables(db) {
       FOREIGN KEY (stash_root, file_path, body_hash)
         REFERENCES graph_files(stash_root, file_path, body_hash) ON DELETE CASCADE
     );
-
-    -- #624-P3: lazy graph-extraction queue. Standalone table (NO FK to
-    -- graph_files — a queued file by definition has no graph row yet).
-    -- Idempotent on (stash_root, file_path); drained highest-priority-first.
-    CREATE TABLE IF NOT EXISTS graph_extraction_queue (
-      stash_root TEXT NOT NULL,
-      file_path  TEXT NOT NULL,
-      body_hash  TEXT NOT NULL,
-      queued_at  TEXT NOT NULL DEFAULT (datetime('now')),
-      priority   INTEGER NOT NULL DEFAULT 0,
-      PRIMARY KEY (stash_root, file_path)
-    );
-
-    CREATE INDEX IF NOT EXISTS idx_graph_extraction_queue_drain
-      ON graph_extraction_queue(stash_root, priority DESC, queued_at);
   `);
 }
 /**
@@ -286,12 +276,17 @@ export function ensureSchema(db) {
     replaceSearchTextWithHash(db);
     // Retired derived tables: the workflow IR cache, the pre-v22 FTS dirty
     // queue, the #955 embedding salvage staging table (embeddings now carry
-    // their model per row, so nothing is copied aside and reused), and the
-    // fragment FTS table search stopped reading (layout 24 and earlier).
+    // their model per row, so nothing is copied aside and reused), the
+    // fragment FTS table search stopped reading (layout 24 and earlier), the
+    // per-project scoped utility table (IR-7a: it shipped, but no code ever read
+    // or wrote a row), and the lazy graph-extraction queue (extraction runs only
+    // in improve).
     db.exec("DROP TABLE IF EXISTS workflow_documents");
     db.exec("DROP TABLE IF EXISTS entries_fts_dirty");
     db.exec("DROP TABLE IF EXISTS embedding_salvage");
     db.exec("DROP TABLE IF EXISTS entry_fragments_fts");
+    db.exec("DROP TABLE IF EXISTS utility_scores_scoped");
+    db.exec("DROP TABLE IF EXISTS graph_extraction_queue");
     // One float32 BLOB per entry, searched by an exact scan
     // (index-vec-repository.ts). `model` is the provider fingerprint the vector was generated under
     // (`deriveSemanticProviderFingerprint`); the embedding pass re-embeds only
@@ -326,19 +321,6 @@ export function ensureSchema(db) {
       updated_at   TEXT NOT NULL DEFAULT (datetime('now')),
       FOREIGN KEY (entry_id) REFERENCES entries(id) ON DELETE CASCADE
     );
-  `);
-    // Per-project scoped utility scores — tracks usage per (entry, cwd-anchor)
-    // so assets useful in project A don't pollute rankings in project B.
-    db.exec(`
-    CREATE TABLE IF NOT EXISTS utility_scores_scoped (
-      entry_id     INTEGER NOT NULL,
-      scope_key    TEXT NOT NULL,
-      utility      REAL NOT NULL DEFAULT 0,
-      last_used_at INTEGER NOT NULL,
-      PRIMARY KEY (entry_id, scope_key)
-    );
-    CREATE INDEX IF NOT EXISTS idx_utility_scores_scoped_entry_id
-      ON utility_scores_scoped(entry_id);
   `);
     db.exec(`
     CREATE TABLE IF NOT EXISTS index_dir_state (

package/docs/reference/cli.md

@@ -503,7 +503,9 @@ embedded with the model's query template (see `embedding.queryTemplate` in
 keyword ranking alone with `fts-fallback` and a warning. Filters (`--type`,
 `--from`, `--filter`, `--belief`, the default session exclusion, proposed
 quality) and one-hit-per-file deduplication narrow the fused list without
-reordering it.
+reordering it. Of entries with identical indexed content (the same body
+saved under another name or in another bundle), only the highest-ranked is
+kept.
 
 | Flag | Values | Default | Description |
 | --- | --- | --- | --- |
@@ -607,6 +609,12 @@ only for read-only items. Their `followUp` remains `akm show <ref>` rather than
 being replaced by clone guidance.
 Use `--type workflow` when you want curated step-by-step procedures instead of
 individual scripts, skills, or docs.
+Curate returns no items, on purpose, when the input is not a task: a harness
+or tool envelope (input that starts with an XML-style tag and contains a
+closing tag, such as `<task-notification>…</task-notification>`) or the stash
+README boilerplate. The `summary` then starts with `Curate abstained` and
+names the reason, and `tip` says how to curate the task instead. Long input is
+curated like any other.
 `akm curate` is safe to call frequently, including from a hook that fires on
 every prompt: it only ever reads the index as it currently stands (the same
 non-blocking `ensureIndex()` path `search` uses) and never waits on or

package/docs/reference/configuration.md

@@ -327,6 +327,13 @@ incompatible engine never falls back to another engine. Built-in strategies
 are complete presets. User-defined strategies inherit omitted fields from the
 built-in `default` strategy before applying their own overrides.
 
+Graph extraction also reads `index.graph`. Its `engine`, `model`, `timeoutMs`
+and `llm` apply over the strategy-wide values, and `processes.graphExtraction`
+overrides them. `processes.graphExtraction.batchSize` and `includeTypes` win
+over `index.graph.graphExtractionBatchSize` and `graphExtractionIncludeTypes`;
+with neither set, the batch size is 4 and the types are `memory` and
+`knowledge`.
+
 `processes.triage.judgment` explicitly controls the optional judgment tier.
 Use `true` to enable it, `false` to disable it, or an object with `enabled`,
 `engine`, `model`, `timeoutMs`, and/or `llm` overrides. Existing object values
@@ -840,3 +847,8 @@ profile identities.
 `embedding.chunkSize` was never read by anything under `src/` (#954), so a
 config that still sets it is simply ignored — it still loads, unvalidated
 and without warning.
+
+`index.graph.lazyGraphExtraction` is retired in 0.9.17: `akm show` and
+`akm curate` no longer extract or queue graph work, and graph extraction runs
+only in `akm improve`. A config that still sets it loads; the key is named
+once as unknown, and `akm migrate apply` removes it.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "akm-cli",
-  "version": "0.9.17-alpha.4",
+  "version": "0.9.17-alpha.6",
   "type": "module",
   "description": "akm (Agent Knowledge Manager) — a portable, local-first capability library for AI agents. Discover, load, share, and improve reusable skills, scripts, workflows, and knowledge across any shell-capable coding agent, including Claude Code, OpenCode, and Cursor.",
   "keywords": [

package/schemas/akm-config.json

@@ -413,9 +413,6 @@
               "minLength": 1
             },
             "minItems": 1
-          },
-          "lazyGraphExtraction": {
-            "type": "boolean"
           }
         },
         "additionalProperties": true
@@ -2077,9 +2074,6 @@
                   "minLength": 1
                 },
                 "minItems": 1
-              },
-              "lazyGraphExtraction": {
-                "type": "boolean"
               }
             },
             "additionalProperties": true