nlm-memory 0.20.0 → 0.21.0

package/migrations/025_workstreams.sql

@@ -0,0 +1,30 @@
+-- migrations/025_workstreams.sql
+-- Workstream abstraction (#367) Plan A: a persistent container a session binds
+-- to at end-of-session. See docs/superpowers/specs/2026-06-24-workstream-abstraction-design.md.
+
+CREATE TABLE IF NOT EXISTS workstreams (
+  id              TEXT PRIMARY KEY,           -- ws_<uuid>
+  label           TEXT NOT NULL,
+  status          TEXT NOT NULL DEFAULT 'active' CHECK(status IN ('active','merged','retired')),
+  merged_into     TEXT REFERENCES workstreams(id),
+  created_at      TEXT NOT NULL DEFAULT (datetime('now')),
+  updated_at      TEXT NOT NULL DEFAULT (datetime('now')),
+  last_session_at TEXT
+);
+
+CREATE TABLE IF NOT EXISTS workstream_entities (
+  workstream_id    TEXT NOT NULL REFERENCES workstreams(id) ON DELETE CASCADE,
+  entity_canonical TEXT NOT NULL REFERENCES entities(canonical),
+  session_count    INTEGER NOT NULL DEFAULT 0,
+  PRIMARY KEY (workstream_id, entity_canonical)
+);
+
+CREATE INDEX IF NOT EXISTS idx_workstream_entities_entity ON workstream_entities(entity_canonical);
+
+ALTER TABLE sessions ADD COLUMN workstream_id TEXT REFERENCES workstreams(id);
+ALTER TABLE sessions ADD COLUMN binding_source TEXT;
+ALTER TABLE sessions ADD COLUMN binding_confidence REAL;
+
+CREATE INDEX IF NOT EXISTS idx_sessions_workstream ON sessions(workstream_id);
+
+INSERT OR IGNORE INTO schema_migrations (version, name) VALUES (25, 'workstreams');

package/migrations/026_embedding_config.sql

@@ -0,0 +1,12 @@
+-- Records which embedder produced the stored vectors, per lane.
+-- lane 'prose' covers session chunks and fact embeddings (same embedder);
+-- lane 'code' covers code exemplar embeddings.
+-- Absence of a row means pre-tracking corpus: reconcile treats the running
+-- embedder as canonical and records it.
+CREATE TABLE IF NOT EXISTS embedding_config (
+  lane       TEXT PRIMARY KEY CHECK (lane IN ('prose', 'code')),
+  provider   TEXT NOT NULL,
+  model      TEXT NOT NULL,
+  dim        INTEGER NOT NULL,
+  updated_at TEXT NOT NULL
+);

package/migrations/027_drop_legacy_session_embeddings.sql

@@ -0,0 +1,7 @@
+-- Migration 027: drop the legacy session_embeddings virtual table.
+-- Migration 003 created it; migration 009 superseded it with the chunk + max-pool index.
+
+DROP TABLE IF EXISTS session_embeddings;
+
+INSERT OR IGNORE INTO schema_migrations (version, name)
+  VALUES (27, '027_drop_legacy_session_embeddings');

package/migrations/028_classifier_provenance.sql

@@ -0,0 +1,5 @@
+-- Which classifier produced this session's classification. NULL = classified
+-- before provenance tracking; nlm reprocess treats NULL as eligible.
+ALTER TABLE sessions ADD COLUMN classifier_provider TEXT;
+ALTER TABLE sessions ADD COLUMN classifier_model TEXT;
+ALTER TABLE sessions ADD COLUMN classifier_confidence REAL;

package/migrations/029_project_scope.sql

@@ -0,0 +1,21 @@
+-- migrations/029_project_scope.sql
+-- Additive scope column on the five corpus tables (#348 Stage A).
+-- No backfill; all existing rows are NULL. Enforcement is deferred behind
+-- NLM_SCOPE_ENFORCE. See docs/superpowers/specs/2026-07-03-project-scoping-design.md.
+
+ALTER TABLE sessions ADD COLUMN scope TEXT;
+CREATE INDEX IF NOT EXISTS idx_sessions_scope ON sessions(scope);
+
+ALTER TABLE facts ADD COLUMN scope TEXT;
+CREATE INDEX IF NOT EXISTS idx_facts_scope ON facts(scope);
+
+ALTER TABLE code_exemplars ADD COLUMN scope TEXT;
+CREATE INDEX IF NOT EXISTS idx_code_exemplars_scope ON code_exemplars(scope);
+
+ALTER TABLE signals ADD COLUMN scope TEXT;
+CREATE INDEX IF NOT EXISTS idx_signals_scope ON signals(scope);
+
+ALTER TABLE workstreams ADD COLUMN scope TEXT;
+CREATE INDEX IF NOT EXISTS idx_workstreams_scope ON workstreams(scope);
+
+INSERT OR IGNORE INTO schema_migrations (version, name) VALUES (29, 'project_scope');

package/migrations/032_session_derivables.sql

@@ -0,0 +1,11 @@
+-- migrations/032_session_derivables.sql
+-- #352 Phase 2: derivable metadata. All nullable: NULL means not-derivable
+-- for that runtime, never faked.
+ALTER TABLE sessions ADD COLUMN agent_persona TEXT;
+ALTER TABLE sessions ADD COLUMN parent_session_id TEXT;
+ALTER TABLE sessions ADD COLUMN primary_model TEXT;
+ALTER TABLE sessions ADD COLUMN total_tokens INTEGER;
+ALTER TABLE sessions ADD COLUMN skill TEXT;
+CREATE INDEX IF NOT EXISTS idx_sessions_parent ON sessions(parent_session_id);
+
+INSERT OR IGNORE INTO schema_migrations (version, name) VALUES (32, 'session_derivables');

package/migrations/pg/001_initial.sql

@@ -26,6 +26,9 @@ CREATE TABLE IF NOT EXISTS sessions (
   transcript_length    BIGINT,
   created_at           TIMESTAMPTZ NOT NULL DEFAULT NOW(),
   updated_at           TIMESTAMPTZ NOT NULL DEFAULT NOW(),
+  workstream_id        TEXT,
+  binding_source       TEXT,
+  binding_confidence   DOUBLE PRECISION,
   fts_vector           TSVECTOR GENERATED ALWAYS AS (
     setweight(to_tsvector('english', coalesce(label, '')), 'A') ||
     setweight(to_tsvector('english', coalesce(summary, '')), 'B') ||
@@ -255,3 +258,34 @@ CREATE TABLE IF NOT EXISTS code_exemplar_embeddings (
 CREATE INDEX IF NOT EXISTS code_exemplar_embeddings_idx
   ON code_exemplar_embeddings USING ivfflat (embedding vector_l2_ops)
   WITH (lists = 100);
+
+-- ── Workstreams (mirror of SQLite migration 025) ─────────────────────────────
+CREATE TABLE IF NOT EXISTS workstreams (
+  id              TEXT PRIMARY KEY,
+  label           TEXT NOT NULL,
+  status          TEXT NOT NULL DEFAULT 'active' CHECK (status IN ('active','merged','retired')),
+  merged_into     TEXT REFERENCES workstreams(id),
+  created_at      TIMESTAMPTZ NOT NULL DEFAULT NOW(),
+  updated_at      TIMESTAMPTZ NOT NULL DEFAULT NOW(),
+  last_session_at TIMESTAMPTZ
+);
+
+CREATE TABLE IF NOT EXISTS workstream_entities (
+  workstream_id    TEXT NOT NULL REFERENCES workstreams(id) ON DELETE CASCADE,
+  entity_canonical TEXT NOT NULL REFERENCES entities(canonical),
+  session_count    INTEGER NOT NULL DEFAULT 0,
+  PRIMARY KEY (workstream_id, entity_canonical)
+);
+
+CREATE INDEX IF NOT EXISTS idx_workstream_entities_entity ON workstream_entities(entity_canonical);
+
+-- FK from sessions.workstream_id to workstreams (deferred so workstreams exists first)
+DO $$
+BEGIN
+  ALTER TABLE sessions
+    ADD CONSTRAINT fk_sessions_workstream
+    FOREIGN KEY (workstream_id) REFERENCES workstreams(id);
+EXCEPTION WHEN duplicate_object THEN NULL;
+END $$;
+
+CREATE INDEX IF NOT EXISTS idx_sessions_workstream ON sessions(workstream_id);

package/migrations/pg/019_split_replaces.sql

@@ -1,9 +1,10 @@
 -- PG parity for SQLite migration 019: split the mechanical `replaces` relation
 -- out of `supersedes`.
 --
--- One-shot repair, applied manually by an operator against a PG canonical store
--- (PgStorage.init only runs 001_initial.sql; there is no version-gated runner on
--- the PG side). Mirrors migrations/019_split_replaces.sql.
+-- Idempotent: safe to re-apply on a database that already has this shape.
+-- DROP CONSTRAINT IF EXISTS removes any existing named constraint before the
+-- DO-block re-adds it, so duplicate_object is never raised in practice;
+-- the EXCEPTION clause is defensive for parity with the 001 pattern.
 --
 -- Widens the CHECK constraints on sessions.status and session_edges.kind to
 -- admit 'replaced' / 'replaces', then reclassifies existing mechanical edges:
@@ -13,12 +14,20 @@
 BEGIN;
 
 ALTER TABLE sessions DROP CONSTRAINT IF EXISTS sessions_status_check;
-ALTER TABLE sessions ADD CONSTRAINT sessions_status_check
-  CHECK (status IN ('active', 'closed', 'superseded', 'replaced'));
+DO $$
+BEGIN
+  ALTER TABLE sessions ADD CONSTRAINT sessions_status_check
+    CHECK (status IN ('active', 'closed', 'superseded', 'replaced'));
+EXCEPTION WHEN duplicate_object THEN NULL;
+END $$;
 
 ALTER TABLE session_edges DROP CONSTRAINT IF EXISTS session_edges_kind_check;
-ALTER TABLE session_edges ADD CONSTRAINT session_edges_kind_check
-  CHECK (kind IN ('supersedes', 'replaces', 'continues'));
+DO $$
+BEGIN
+  ALTER TABLE session_edges ADD CONSTRAINT session_edges_kind_check
+    CHECK (kind IN ('supersedes', 'replaces', 'continues'));
+EXCEPTION WHEN duplicate_object THEN NULL;
+END $$;
 
 UPDATE session_edges e SET kind = 'replaces'
 WHERE e.kind = 'supersedes'

package/migrations/pg/025_workstreams.sql

@@ -0,0 +1,34 @@
+-- migrations/pg/025_workstreams.sql
+-- PG parity for SQLite migration 025 (workstreams). Auto-applied by the
+-- version-gated runner (runMigrationsPg). Idempotent: every DDL statement uses
+-- IF NOT EXISTS or ADD COLUMN IF NOT EXISTS, so re-applying on an existing
+-- schema is a no-op. Mirrors migrations/025_workstreams.sql.
+
+BEGIN;
+
+CREATE TABLE IF NOT EXISTS workstreams (
+  id              TEXT PRIMARY KEY,
+  label           TEXT NOT NULL,
+  status          TEXT NOT NULL DEFAULT 'active' CHECK (status IN ('active','merged','retired')),
+  merged_into     TEXT REFERENCES workstreams(id),
+  created_at      TIMESTAMPTZ NOT NULL DEFAULT NOW(),
+  updated_at      TIMESTAMPTZ NOT NULL DEFAULT NOW(),
+  last_session_at TIMESTAMPTZ
+);
+
+CREATE TABLE IF NOT EXISTS workstream_entities (
+  workstream_id    TEXT NOT NULL REFERENCES workstreams(id) ON DELETE CASCADE,
+  entity_canonical TEXT NOT NULL REFERENCES entities(canonical),
+  session_count    INTEGER NOT NULL DEFAULT 0,
+  PRIMARY KEY (workstream_id, entity_canonical)
+);
+
+CREATE INDEX IF NOT EXISTS idx_workstream_entities_entity ON workstream_entities(entity_canonical);
+
+ALTER TABLE sessions ADD COLUMN IF NOT EXISTS workstream_id TEXT REFERENCES workstreams(id);
+ALTER TABLE sessions ADD COLUMN IF NOT EXISTS binding_source TEXT;
+ALTER TABLE sessions ADD COLUMN IF NOT EXISTS binding_confidence DOUBLE PRECISION;
+
+CREATE INDEX IF NOT EXISTS idx_sessions_workstream ON sessions(workstream_id);
+
+COMMIT;

package/migrations/pg/026_embedding_config.sql

@@ -0,0 +1,12 @@
+-- Records which embedder produced the stored vectors, per lane.
+-- lane 'prose' covers session chunks and fact embeddings (same embedder);
+-- lane 'code' covers code exemplar embeddings.
+-- Absence of a row means pre-tracking corpus: reconcile treats the running
+-- embedder as canonical and records it.
+CREATE TABLE IF NOT EXISTS embedding_config (
+  lane       TEXT PRIMARY KEY CHECK (lane IN ('prose', 'code')),
+  provider   TEXT NOT NULL,
+  model      TEXT NOT NULL,
+  dim        INTEGER NOT NULL,
+  updated_at TEXT NOT NULL
+);

package/migrations/pg/028_classifier_provenance.sql

@@ -0,0 +1,5 @@
+-- Which classifier produced this session's classification. NULL = classified
+-- before provenance tracking; nlm reprocess treats NULL as eligible.
+ALTER TABLE sessions ADD COLUMN IF NOT EXISTS classifier_provider TEXT;
+ALTER TABLE sessions ADD COLUMN IF NOT EXISTS classifier_model TEXT;
+ALTER TABLE sessions ADD COLUMN IF NOT EXISTS classifier_confidence REAL;

package/migrations/pg/029_entity_variants.sql

@@ -0,0 +1,9 @@
+-- Entity variant alias table. Tracks merged surface forms so re-ingest of an
+-- old spelling binds to the canonical entity instead of resurrecting the source.
+CREATE TABLE IF NOT EXISTS entity_variants (
+  variant           TEXT PRIMARY KEY,
+  canonical         TEXT NOT NULL REFERENCES entities(canonical) ON DELETE CASCADE,
+  source_session_id TEXT REFERENCES sessions(id) ON DELETE SET NULL
+);
+
+CREATE INDEX IF NOT EXISTS idx_entity_variants_canonical ON entity_variants(canonical);

package/migrations/pg/030_project_scope.sql

@@ -0,0 +1,23 @@
+-- migrations/pg/030_project_scope.sql
+-- PG parity for SQLite migration 029 (project scope). Auto-applied by the
+-- version-gated runner (runMigrationsPg). Idempotent: every DDL uses
+-- ADD COLUMN IF NOT EXISTS and CREATE INDEX IF NOT EXISTS. Mirrors migrations/029_project_scope.sql.
+
+BEGIN;
+
+ALTER TABLE sessions ADD COLUMN IF NOT EXISTS scope TEXT;
+CREATE INDEX IF NOT EXISTS idx_sessions_scope ON sessions(scope);
+
+ALTER TABLE facts ADD COLUMN IF NOT EXISTS scope TEXT;
+CREATE INDEX IF NOT EXISTS idx_facts_scope ON facts(scope);
+
+ALTER TABLE code_exemplars ADD COLUMN IF NOT EXISTS scope TEXT;
+CREATE INDEX IF NOT EXISTS idx_code_exemplars_scope ON code_exemplars(scope);
+
+ALTER TABLE signals ADD COLUMN IF NOT EXISTS scope TEXT;
+CREATE INDEX IF NOT EXISTS idx_signals_scope ON signals(scope);
+
+ALTER TABLE workstreams ADD COLUMN IF NOT EXISTS scope TEXT;
+CREATE INDEX IF NOT EXISTS idx_workstreams_scope ON workstreams(scope);
+
+COMMIT;

package/migrations/pg/031_widen_session_edges_kind.sql

@@ -0,0 +1,11 @@
+-- Parity: SQLite migrations/019 allows five session_edges.kind values
+-- (supersedes, replaces, continues, branched_from, merged_from); pg/019
+-- only widened to three. This closes the remaining gap.
+
+BEGIN;
+
+ALTER TABLE session_edges DROP CONSTRAINT IF EXISTS session_edges_kind_check;
+ALTER TABLE session_edges ADD CONSTRAINT session_edges_kind_check
+  CHECK (kind IN ('supersedes', 'replaces', 'continues', 'branched_from', 'merged_from'));
+
+COMMIT;

package/migrations/pg/032_session_derivables.sql

@@ -0,0 +1,17 @@
+-- migrations/pg/032_session_derivables.sql
+-- PG parity for SQLite migration 032 (session derivables). Auto-applied by
+-- the version-gated runner (runMigrationsPg). Idempotent: every DDL uses
+-- ADD COLUMN IF NOT EXISTS and CREATE INDEX IF NOT EXISTS. Mirrors
+-- migrations/032_session_derivables.sql. All nullable: NULL means
+-- not-derivable for that runtime, never faked.
+
+BEGIN;
+
+ALTER TABLE sessions ADD COLUMN IF NOT EXISTS agent_persona TEXT;
+ALTER TABLE sessions ADD COLUMN IF NOT EXISTS parent_session_id TEXT;
+ALTER TABLE sessions ADD COLUMN IF NOT EXISTS primary_model TEXT;
+ALTER TABLE sessions ADD COLUMN IF NOT EXISTS total_tokens INTEGER;
+ALTER TABLE sessions ADD COLUMN IF NOT EXISTS skill TEXT;
+CREATE INDEX IF NOT EXISTS idx_sessions_parent ON sessions(parent_session_id);
+
+COMMIT;