talon-agent 1.9.0 → 1.9.2

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "talon-agent",
-  "version": "1.9.0",
+  "version": "1.9.2",
   "description": "Multi-frontend AI agent with full tool access, streaming, cron jobs, and plugin system",
   "author": "Dylan Neve",
   "license": "MIT",
@@ -51,13 +51,14 @@
   },
   "dependencies": {
     "@anthropic-ai/claude-agent-sdk": "^0.2.108",
+    "@anthropic-ai/sdk": "^0.95.0",
     "@brave/brave-search-mcp-server": "^2.0.75",
     "@clack/prompts": "^1.2.0",
     "@grammyjs/auto-retry": "^2.0.2",
     "@grammyjs/transformer-throttler": "^1.2.1",
     "@modelcontextprotocol/sdk": "^1.29.0",
     "@opencode-ai/sdk": "^1.4.0",
-    "@playwright/mcp": "^0.0.70",
+    "@playwright/mcp": "^0.0.74",
     "big-integer": "^1.6.52",
     "cheerio": "^1.2.0",
     "croner": "^10.0.1",
@@ -85,6 +86,6 @@
     "vitest": "^4.1.3"
   },
   "overrides": {
-    "@anthropic-ai/sdk": "^0.86.1"
+    "@anthropic-ai/sdk": "^0.95.0"
   }
 }

package/prompts/mempalace.md

@@ -7,6 +7,7 @@ You have access to a local memory palace via MCP tools. The palace stores verbat
 - **Wings** = top-level categories (people, projects, topics)
 - **Rooms** = specific subjects within a wing
 - **Drawers** = individual memory chunks (verbatim text)
+- **Tunnels** = cross-wing links between related rooms (auto-created in mempalace 3.3.4+ when topics overlap, plus manual)
 - **Knowledge Graph** = entity-relationship facts with temporal validity
 
 ### Protocol — FOLLOW EVERY SESSION
@@ -25,6 +26,7 @@ You have access to a local memory palace via MCP tools. The palace stores verbat
 - `mempalace_status` — Palace overview: total drawers, wings, rooms.
 - `mempalace_list_wings` / `mempalace_list_rooms` — Browse structure.
 - `mempalace_get_taxonomy` — Full wing/room/count tree.
+- `mempalace_get_aaak_spec` — Get the AAAK closet/compression spec. Only needed when reading/writing AAAK-compressed memories directly.
 
 **Knowledge Graph (Temporal Facts):**
 
@@ -34,24 +36,42 @@ You have access to a local memory palace via MCP tools. The palace stores verbat
 - `mempalace_kg_timeline` — Chronological story of an entity.
 - `mempalace_kg_stats` — Graph overview: entities, triples, relationship types.
 
-**Palace Graph (Cross-Domain Connections):**
+**Palace Graph & Cross-Wing Tunnels:**
 
 - `mempalace_traverse` — Walk from a room, find connected ideas across wings.
 - `mempalace_find_tunnels` — Find rooms that bridge two wings.
+- `mempalace_follow_tunnels` — From a specific (wing, room) pair, walk the outbound tunnels and see connected rooms with drawer previews.
+- `mempalace_create_tunnel` — Manually create a cross-wing tunnel between two (wing, room) pairs. Use when you spot a connection the auto-detector missed.
+- `mempalace_list_tunnels` — List tunnels (optionally filtered by wing).
+- `mempalace_delete_tunnel` — Remove a tunnel by ID when it's wrong or noisy.
 - `mempalace_graph_stats` — Graph connectivity overview.
 
-**Write:**
+**Drawers:**
 
 - `mempalace_add_drawer` — Store verbatim content into a wing/room. Auto-checks duplicates.
+- `mempalace_get_drawer` — Fetch a single drawer by ID. Returns full content + metadata. Use after a search hit when you need the verbatim text.
+- `mempalace_list_drawers` — Browse drawers in a wing/room with pagination (`limit`, `offset`). Use for inventory/cleanup, not search.
+- `mempalace_update_drawer` — Edit an existing drawer's content, wing, or room in place. Use to refine misfiled or stale entries instead of delete + re-add.
 - `mempalace_delete_drawer` — Remove a drawer by ID.
-- `mempalace_diary_write` — Write a session diary entry (agent_name, entry, topic).
-- `mempalace_diary_read` — Read recent diary entries.
+
+**Diary:**
+
+- `mempalace_diary_write` — Write a session diary entry (`agent_name`, `entry`, `topic`, optional `wing`).
+- `mempalace_diary_read` — Read recent diary entries. Optional `wing` filter scopes by project.
+
+**Maintenance:**
+
+- `mempalace_memories_filed_away` — Check whether a recent checkpoint was saved (message count, timestamp). Useful for confirming a stop-hook flush happened.
+- `mempalace_reconnect` — Force reconnect to the palace database. Run after external CLI/scripts modify the palace directly (the in-memory HNSW index can otherwise go stale).
+- `mempalace_hook_settings` — Toggle silent-save / desktop-toast for the auto-save hooks.
 
 ### Tips
 
 - Search is **semantic** (meaning-based), not keyword. "What did we discuss about database performance?" works better than "database".
 - The knowledge graph stores typed relationships with **time windows**. It knows WHEN things were true.
 - Use `mempalace_check_duplicate` before storing new content to avoid clutter.
+- **Tunnels auto-form** when drawers across different wings share topics (mempalace 3.3.4+). You don't have to wire connections by hand most of the time — but `mempalace_create_tunnel` is there when the auto-detector misses something obvious, and `mempalace_delete_tunnel` is there when it overreaches.
+- After updating facts via the CLI / external scripts, call `mempalace_reconnect` so the live MCP server picks up the changes.
 - Diary entries accumulate across sessions. Write them to build continuity of self.
 - Entity detection runs per-language; results include `created_at` timestamps you can surface when the user asks "when did I last…".
 

package/src/__tests__/disallowed-tools.test.ts

@@ -0,0 +1,64 @@
+import { describe, expect, it } from "vitest";
+
+import {
+  DISALLOWED_TOOLS_BACKGROUND,
+  DISALLOWED_TOOLS_CORE,
+} from "../core/constants.js";
+import { DISALLOWED_TOOLS_CHAT } from "../backend/claude-sdk/constants.js";
+
+describe("disallowed tool lists", () => {
+  describe("DISALLOWED_TOOLS_CORE", () => {
+    it("blocks interactive/planning tools that are nonsensical in headless contexts", () => {
+      const expected = [
+        "EnterPlanMode",
+        "ExitPlanMode",
+        "EnterWorktree",
+        "ExitWorktree",
+        "TodoWrite",
+        "TodoRead",
+        "TaskCreate",
+        "TaskUpdate",
+        "TaskGet",
+        "TaskList",
+        "TaskOutput",
+        "TaskStop",
+        "AskUserQuestion",
+      ];
+      for (const tool of expected) {
+        expect(DISALLOWED_TOOLS_CORE).toContain(tool);
+      }
+    });
+
+    it("blocks ScheduleWakeup — /loop-skill-only tool that wedges the dispatcher when called outside /loop mode", () => {
+      // Confirmed root cause of a 35-minute hang on 2026-04-27.
+      // ScheduleWakeup registers a wakeup the runtime never fires, so the
+      // chat lock is held indefinitely until manual restart.
+      expect(DISALLOWED_TOOLS_CORE).toContain("ScheduleWakeup");
+    });
+  });
+
+  describe("DISALLOWED_TOOLS_BACKGROUND", () => {
+    it("inherits everything from CORE", () => {
+      for (const tool of DISALLOWED_TOOLS_CORE) {
+        expect(DISALLOWED_TOOLS_BACKGROUND).toContain(tool);
+      }
+    });
+
+    it("additionally blocks Agent (no nested agents in dream/heartbeat)", () => {
+      expect(DISALLOWED_TOOLS_BACKGROUND).toContain("Agent");
+    });
+  });
+
+  describe("DISALLOWED_TOOLS_CHAT", () => {
+    it("inherits everything from CORE", () => {
+      for (const tool of DISALLOWED_TOOLS_CORE) {
+        expect(DISALLOWED_TOOLS_CHAT).toContain(tool);
+      }
+    });
+
+    it("additionally blocks Claude's built-in web tools (replaced by Brave MCP)", () => {
+      expect(DISALLOWED_TOOLS_CHAT).toContain("WebSearch");
+      expect(DISALLOWED_TOOLS_CHAT).toContain("WebFetch");
+    });
+  });
+});

package/src/__tests__/handlers-stream.test.ts

@@ -29,11 +29,6 @@ vi.mock("../core/errors.js", () => ({
   TalonError: class TalonError extends Error {},
 }));
 
-vi.mock("../core/prompt-builder.js", () => ({
-  enrichDMPrompt: vi.fn((p: string) => p),
-  enrichGroupPrompt: vi.fn((p: string) => p),
-}));
-
 vi.mock("../storage/daily-log.js", () => ({
   appendDailyLog: vi.fn(),
   appendDailyLogResponse: vi.fn(),

package/src/__tests__/handlers.test.ts

@@ -5,10 +5,6 @@ const executeMock = vi.hoisted(() => vi.fn());
 vi.mock("../core/dispatcher.js", () => ({
   execute: executeMock,
 }));
-vi.mock("../core/prompt-builder.js", () => ({
-  enrichDMPrompt: vi.fn((p: string) => p),
-  enrichGroupPrompt: vi.fn((p: string) => p),
-}));
 vi.mock("../storage/daily-log.js", () => ({
   appendDailyLog: vi.fn(),
   appendDailyLogResponse: vi.fn(),
@@ -2641,11 +2637,8 @@ describe("handleStickerMessage — video sticker branch (L835 TRUE)", () => {
   }, 3000);
 });
 
-describe("processAndReply — group message without senderId (L552 FALSE branch)", () => {
-  it("skips enrichGroupPrompt when senderId is undefined in group", async () => {
-    const { enrichGroupPrompt } = await import("../core/prompt-builder.js");
-    (enrichGroupPrompt as ReturnType<typeof vi.fn>).mockClear();
-
+describe("processAndReply — group message without senderId", () => {
+  it("processes anonymous group messages without mutating the prompt", async () => {
     executeMock.mockResolvedValueOnce({
       text: "",
       durationMs: 10,
@@ -2672,13 +2665,14 @@ describe("processAndReply — group message without senderId (L552 FALSE branch)
     await handleTextMessage(ctx, mockBot, mockConfig);
     await new Promise((r) => setTimeout(r, 700));
 
-    // enrichGroupPrompt should NOT have been called (senderId falsy)
-    expect(enrichGroupPrompt).not.toHaveBeenCalled();
-    // But message was still processed
+    // Message was still processed, and the prompt is passed through verbatim.
     const calls = executeMock.mock.calls
       .slice(before)
       .filter((c) => (c[0] as { chatId: string }).chatId === String(chatId));
     expect(calls.length).toBe(1);
+    expect((calls[0][0] as { prompt: string }).prompt).toBe(
+      "@testbot anonymous message",
+    );
   }, 3000);
 });
 

package/src/__tests__/log.test.ts

@@ -71,10 +71,11 @@ describe("log", () => {
       );
     });
 
-    it("includes Error message in context", () => {
-      logError("bridge", "request failed", new Error("timeout"));
+    it("includes Error message and stack in context", () => {
+      const err = new Error("timeout");
+      logError("bridge", "request failed", err);
       expect(mockError).toHaveBeenCalledWith(
-        { component: "bridge", err: "timeout" },
+        { component: "bridge", err: "timeout", stack: err.stack },
         "request failed",
       );
     });