osborn 0.9.220 → 0.9.222

package/dist/claude-llm.js

@@ -86,13 +86,13 @@ async function buildRecallInjection(sessionId, workingDir, prompt) {
         return ''; // recall is best-effort — never break the turn
     }
 }
-// ≤3 direct tool call budget per turn. Reset on every UserPromptSubmit (new user message).
+// ≤5 direct tool call budget per turn. Reset on every UserPromptSubmit (new user message).
 // Enforced mechanically in PreToolUse — the model CANNOT exceed this regardless of JSONL history.
 // Task/Agent delegations are exempt (delegation is what we WANT). Sub-agent tool calls
 // (agent_type !== null) are exempt (they're inside a delegation). Only the main orchestrator
 // agent's direct tool calls count against the budget.
 let turnToolCallCount = 0;
-const TOOL_CALL_BUDGET = 3;
+const TOOL_CALL_BUDGET = 5;
 /**
  * Strip markdown formatting for TTS (text-to-speech)
  * Removes **bold**, ##headers, ```code```, etc. so TTS doesn't read them literally
@@ -1704,6 +1704,17 @@ export class ClaudeLLM extends llm.LLM {
                                     return {};
                                 }],
                         }],
+                    // Self-review gate — block ONCE (stop_hook_active-guarded) so the reviewer
+                    // re-checks its own verdict before returning, instead of stopping on a
+                    // first-pass judgment.
+                    Stop: [{
+                            matcher: '.*',
+                            hooks: [async (input) => {
+                                    if (input?.stop_hook_active)
+                                        return {};
+                                    return { decision: 'block', reason: 'Before you finalize: re-read your findings against the diff once. Confirm each BLOCKER/MAJOR is real and reproducible (not a style nit or a false positive), and that your verdict matches the severity of what you actually found. Then end with exactly `VERDICT: ACCEPT` or `VERDICT: REJECT`.' };
+                                }],
+                        }],
                 },
             };
             console.log(`[DISPATCH] spawning reviewer for agentId=${agentId.slice(0, 8)}`);
@@ -1793,6 +1804,16 @@ export class ClaudeLLM extends llm.LLM {
                                     return {};
                                 }],
                         }],
+                    // Self-review gate — block ONCE (stop_hook_active-guarded) so the tester
+                    // re-checks its own conclusion before returning.
+                    Stop: [{
+                            matcher: '.*',
+                            hooks: [async (input) => {
+                                    if (input?.stop_hook_active)
+                                        return {};
+                                    return { decision: 'block', reason: 'Before you finalize: re-check your conclusion once. Did you actually RUN the commands (not assume the outcome)? Does your result match the real output, and did you distinguish a genuine regression from a flaky/environment failure? Then end with exactly `RESULT: PASS` or `RESULT: FAIL` followed by a brief summary.' };
+                                }],
+                        }],
                 },
             };
             console.log(`[DISPATCH] spawning tester for agentId=${agentId.slice(0, 8)}`);
@@ -1879,6 +1900,16 @@ export class ClaudeLLM extends llm.LLM {
                                     return {};
                                 }],
                         }],
+                    // Self-review gate — block ONCE (stop_hook_active-guarded) so the gate
+                    // re-checks its own PASS/NEEDS-MORE call before returning.
+                    Stop: [{
+                            matcher: '.*',
+                            hooks: [async (input) => {
+                                    if (input?.stop_hook_active)
+                                        return {};
+                                    return { decision: 'block', reason: 'Before you finalize: re-check your gate decision once. Is the research genuinely complete and well-sourced for the original question — not passing thin work, nor failing solid work? Then end with exactly `GATE: PASS` or `GATE: NEEDS-MORE`.' };
+                                }],
+                        }],
                 },
             };
             console.log(`[DISPATCH] spawning research-gate for agentId=${agentId.slice(0, 8)}`);

package/dist/embedder.js

@@ -15,8 +15,8 @@
  *
  * Model cache honors TRANSFORMERS_CACHE / HF_HOME so it can be baked into the image.
  */
-import { EMBED_DIM } from './session-store.js';
-const MODEL_ID = process.env.OSBORN_EMBED_MODEL || 'Xenova/all-MiniLM-L6-v2';
+import { EMBED_DIM, EMBED_MODEL } from './session-store.js';
+const MODEL_ID = EMBED_MODEL;
 let pipelinePromise = null;
 let disabled = false;
 /** Quantize a normalized float32 embedding to int8[dim] (×127, clamped). */

package/dist/session-store.d.ts

@@ -30,6 +30,9 @@
 import Database from 'better-sqlite3';
 export declare const STORE_VERSION = 1;
 export declare const EMBED_DIM = 384;
+/** Canonical embed model id — the single source of truth (embedder.ts imports this).
+ * Recorded in meta.embed_model so a model swap is detected as a stale store. */
+export declare const EMBED_MODEL: string;
 /** An embedder turns text into int8[EMBED_DIM] vectors (quantized, normalized). */
 export type Embedder = (texts: string[]) => Promise<Int8Array[] | null>;
 export interface RecallHit {

package/dist/session-store.js

@@ -39,6 +39,9 @@ import { getSessionPaths, getSessionSubAgents, projectPathToSlug } from './sessi
 // ============================================================
 export const STORE_VERSION = 1;
 export const EMBED_DIM = 384;
+/** Canonical embed model id — the single source of truth (embedder.ts imports this).
+ * Recorded in meta.embed_model so a model swap is detected as a stale store. */
+export const EMBED_MODEL = process.env.OSBORN_EMBED_MODEL || 'Xenova/all-MiniLM-L6-v2';
 // ============================================================
 // PATHS
 // ============================================================
@@ -64,6 +67,41 @@ export function openStore(dbPath) {
     sqliteVec.load(db);
     db.pragma('journal_mode = WAL');
     db.pragma('synchronous = NORMAL');
+    // ── Migration gate ──────────────────────────────────────────────
+    // The derived tables (content/fts/vec/sources) are a re-derivable cache of the
+    // JSONL, which is the source of truth. If this DB was built by a different
+    // STORE_VERSION, EMBED_DIM, or embed model, the cache is stale — most fatally the
+    // vec table is fixed at its old dimension, so queries embedded at the new dim throw
+    // or return garbage. Detect that and drop the derived tables + reset offsets; the
+    // next updateSessionStore() rebuilds them fresh from the JSONL at offset 0.
+    //
+    // A store with NO recorded embed_model (i.e. predating this stamp) is treated as
+    // NOT stale — otherwise the first rollout of this code would wipe every existing DB.
+    // We only wipe on a real mismatch, never on absence.
+    const hasMeta = db
+        .prepare(`SELECT 1 FROM sqlite_master WHERE type='table' AND name='meta'`)
+        .get();
+    if (hasMeta) {
+        const readMeta = db.prepare('SELECT value FROM meta WHERE key = ?');
+        const prevVersion = readMeta.get('version')?.value ?? null;
+        const prevDim = readMeta.get('embed_dim')?.value ?? null;
+        const prevModel = readMeta.get('embed_model')?.value ?? null;
+        const stale = (prevVersion != null && prevVersion !== String(STORE_VERSION)) ||
+            (prevDim != null && prevDim !== String(EMBED_DIM)) ||
+            (prevModel != null && prevModel !== EMBED_MODEL);
+        if (stale) {
+            console.warn(`⚠️  recall store ${dbPath} is stale ` +
+                `(version ${prevVersion}→${STORE_VERSION}, dim ${prevDim}→${EMBED_DIM}, ` +
+                `model ${prevModel}→${EMBED_MODEL}) — rebuilding derived tables from JSONL`);
+            db.exec(`
+        DROP TABLE IF EXISTS content;
+        DROP TABLE IF EXISTS fts;
+        DROP TABLE IF EXISTS vec;
+        DROP TABLE IF EXISTS sources;
+        DELETE FROM meta WHERE key IN ('version', 'embed_dim', 'embed_model');
+      `);
+        }
+    }
     db.exec(`
     CREATE TABLE IF NOT EXISTS content (
       id          INTEGER PRIMARY KEY,
@@ -94,9 +132,13 @@ export function openStore(dbPath) {
 
     CREATE TABLE IF NOT EXISTS meta (key TEXT PRIMARY KEY, value TEXT);
   `);
-    const setMeta = db.prepare('INSERT OR IGNORE INTO meta(key, value) VALUES (?, ?)');
+    // INSERT OR REPLACE (not IGNORE) so a rebuilt store restamps its identity; the
+    // migration gate above cleared these rows on a stale open, so this writes the
+    // current version/dim/model that the freshly-created tables were built for.
+    const setMeta = db.prepare('INSERT OR REPLACE INTO meta(key, value) VALUES (?, ?)');
     setMeta.run('version', String(STORE_VERSION));
     setMeta.run('embed_dim', String(EMBED_DIM));
+    setMeta.run('embed_model', EMBED_MODEL);
     return db;
 }
 // ============================================================

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "osborn",
-  "version": "0.9.220",
+  "version": "0.9.222",
   "description": "Voice AI coding assistant - local agent that connects to Osborn frontend",
   "type": "module",
   "bin": {