talon-agent 5.0.0 → 5.1.0

package/src/backend/claude-sdk/one-shot.ts

@@ -20,6 +20,7 @@ import { ALLOWED_TOOLS_BACKGROUND } from "../../core/constants.js";
 import { EFFORT_MAP } from "./constants.js";
 import { buildMcpServers, buildPluginMcpServers } from "./options.js";
 import { warnIfBelowCacheMinimum } from "../runtime/cache/cache-telemetry.js";
+import { emitAssistantText } from "../runtime/one-shot-hooks.js";
 
 const DEFAULT_SUBPROCESS_KILL_GRACE_MS = 5 * 1000;
 
@@ -65,6 +66,7 @@ export async function runOneShotAgent(
     contextLabel,
     abortController,
     appendLog,
+    onAssistantText,
   } = params;
 
   // Reasoning effort is opt-in for background runs (config `heartbeatEffort`
@@ -126,7 +128,7 @@ export async function runOneShotAgent(
   // settlement figure the task table records.
   let usage: OneShotUsage | undefined;
   for await (const msg of qi) {
-    await formatAndAppendMessage(appendLog, msg);
+    await formatAndAppendMessage(appendLog, msg, onAssistantText);
     if (msg.type === "result") {
       const u = msg.usage;
       usage = {
@@ -183,6 +185,7 @@ function assembleMcpServers(contextLabel: string): Record<string, unknown> {
 async function formatAndAppendMessage(
   appendLog: (text: string) => Promise<void>,
   msg: SDKMessage,
+  onAssistantText?: OneShotAgentParams["onAssistantText"],
 ): Promise<void> {
   try {
     const ts = new Date().toISOString().slice(11, 19);
@@ -200,7 +203,11 @@ async function formatAndAppendMessage(
           });
 
         if (textBlocks.length > 0) {
-          await appendLog(`\n## [${ts}] Assistant\n${textBlocks.join("\n")}\n`);
+          const assistantText = textBlocks.join("\n");
+          // Report before the log write: an append failure (full disk, closed
+          // handle) must not also swallow the run's result for a hook caller.
+          emitAssistantText(onAssistantText, assistantText);
+          await appendLog(`\n## [${ts}] Assistant\n${assistantText}\n`);
         }
         if (toolUseBlocks.length > 0) {
           await appendLog(`\n${toolUseBlocks.join("\n\n")}\n`);
@@ -208,6 +215,14 @@ async function formatAndAppendMessage(
         break;
       }
       case "result": {
+        // Deliberately NOT reported through `onAssistantText`. The SDK's
+        // terminal `result` message restates the last assistant turn's text
+        // (`subtype: "success"`) or carries an error string (the
+        // `error_*` subtypes) — never anything the `assistant` case above
+        // has not already emitted. Reporting it too would hand every hook
+        // caller a duplicate final segment, and the truncated copy at that
+        // (2000 chars, below). The log keeps it because a run log wants the
+        // settlement line; a sub-agent result does not.
         const result =
           "result" in msg
             ? (msg as { result: string }).result

package/src/backend/claude-sdk/stream.ts

@@ -422,6 +422,15 @@ function readResultError(msg: SDKResultMessage, state: StreamState): void {
       .filter((e) => typeof e === "string" && !e.startsWith("[ede_diagnostic]"))
       .join("; ")
       .slice(0, 500);
+    // A known startup failure (SDK ≥ 0.3.274) names its cause — the CLI
+    // never ran a turn, so nothing else in the result says why. Lead with
+    // it; the diagnostics text is the same line stderr carried.
+    if (msg.startup_failure_reason) {
+      state.resultErrorText =
+        `Claude Code failed to start (${msg.startup_failure_reason})` +
+        (diagnostics ? `: ${diagnostics}` : "");
+      return;
+    }
     state.resultErrorText =
       state.lastTrailingText.trim() ||
       diagnostics ||

package/src/backend/codex/mcp-config.ts

@@ -29,7 +29,7 @@ import {
   pluginHubUrl,
   hubPluginServerNames,
 } from "../../core/mcp-hub/index.js";
-import type { ToolExclusionConfig } from "../../core/tools/mcp-env.js";
+import type { ToolExclusionConfig } from "../../core/tools/ops/mcp-env.js";
 import { frontendsForChat } from "../runtime/frontends.js";
 
 /**

package/src/backend/codex/one-shot.ts

@@ -20,6 +20,7 @@
 import type { OneShotAgentParams, OneShotUsage } from "../../core/types.js";
 import { log, logWarn } from "../../util/log.js";
 import { appendBackendSuffix } from "../runtime/index.js";
+import { emitAssistantText } from "../runtime/one-shot-hooks.js";
 import { ensureCodex, getCodexAuthInfo } from "./init.js";
 import {
   CODEX_SYSTEM_PROMPT_SUFFIX,
@@ -79,6 +80,7 @@ export async function runOneShotAgent(
     contextLabel,
     abortController,
     appendLog,
+    onAssistantText,
   } = params;
 
   const codex = ensureCodex(contextLabel);
@@ -143,7 +145,7 @@ export async function runOneShotAgent(
     let usage: OneShotUsage | undefined;
     for await (const event of events) {
       if (abortController.signal.aborted) break;
-      await appendCodexEvent(appendLog, event);
+      await appendCodexEvent(appendLog, event, onAssistantText);
       if (event.type === "turn.completed") {
         const u = (event as { usage?: Record<string, number> }).usage;
         if (u) {
@@ -217,6 +219,7 @@ export async function runOneShotAgent(
 async function appendCodexEvent(
   appendLog: (text: string) => Promise<void>,
   event: { type: string } & Record<string, unknown>,
+  onAssistantText?: OneShotAgentParams["onAssistantText"],
 ): Promise<void> {
   const ts = new Date().toISOString().slice(11, 19);
 
@@ -260,7 +263,7 @@ async function appendCodexEvent(
     case "item.completed": {
       const item = (event as unknown as { item?: Record<string, unknown> })
         .item;
-      if (item) await appendCodexItem(appendLog, item, ts);
+      if (item) await appendCodexItem(appendLog, item, ts, onAssistantText);
       return;
     }
     default:
@@ -268,17 +271,28 @@ async function appendCodexEvent(
   }
 }
 
-/** Append one `ThreadItem` to the run log. */
+/**
+ * Append one `ThreadItem` to the run log, and report the model's final
+ * answers to the run's optional `onAssistantText` consumer.
+ *
+ * Only `agent_message` items are reported: `reasoning` items are the model's
+ * thinking and the rest are tool/command/diff payloads, none of which is the
+ * run's answer.
+ */
 async function appendCodexItem(
   appendLog: (text: string) => Promise<void>,
   item: Record<string, unknown>,
   ts: string,
+  onAssistantText?: OneShotAgentParams["onAssistantText"],
 ): Promise<void> {
   const type = typeof item.type === "string" ? item.type : "unknown";
 
   if (type === "agent_message") {
     const text = typeof item.text === "string" ? item.text : "";
-    if (text) await appendLog(`\n## [${ts}] Assistant\n${text}\n`);
+    if (text) {
+      emitAssistantText(onAssistantText, text);
+      await appendLog(`\n## [${ts}] Assistant\n${text}\n`);
+    }
     return;
   }
 

package/src/backend/openai-agents/mcp-pool.ts

@@ -30,7 +30,7 @@ import {
   pluginHubUrl,
   hubPluginServerNames,
 } from "../../core/mcp-hub/index.js";
-import type { ToolExclusionConfig } from "../../core/tools/mcp-env.js";
+import type { ToolExclusionConfig } from "../../core/tools/ops/mcp-env.js";
 import { frontendsForChat } from "../runtime/frontends.js";
 import { log, logWarn } from "../../util/log.js";
 

package/src/backend/remote-server/one-shot.ts

@@ -42,6 +42,7 @@ import {
   type RemoteAssistantInfo,
 } from "./session-helpers.js";
 import { appendBackendSuffix, sleep } from "../runtime/index.js";
+import { emitAssistantText } from "../runtime/one-shot-hooks.js";
 import { buildPermissionRuleset } from "./sessions.js";
 
 // ── Client surface ──────────────────────────────────────────────────────────
@@ -108,6 +109,7 @@ export async function runRemoteOneShotAgent<
     contextLabel,
     abortController,
     appendLog,
+    onAssistantText,
   } = params;
   const { label, errMsg } = bindings;
 
@@ -210,7 +212,7 @@ export async function runRemoteOneShotAgent<
       : [];
 
     for (const part of parts) {
-      await appendResponsePart(appendLog, part);
+      await appendResponsePart(appendLog, part, onAssistantText);
     }
 
     // The prompt response's assistant info carries the run's token usage —
@@ -250,17 +252,28 @@ export async function runRemoteOneShotAgent<
 
 // ── Run-log rendering ───────────────────────────────────────────────────────
 
-/** Render one response part into the Markdown run log. */
+/**
+ * Render one response part into the Markdown run log, and report the
+ * assistant's final text to the run's optional `onAssistantText` consumer.
+ *
+ * Only the `text` parts are reported: `reasoning` parts are the model's
+ * thinking and tool parts are call payloads, neither of which is the run's
+ * answer.
+ */
 async function appendResponsePart(
   appendLog: (text: string) => Promise<void>,
   part: Record<string, unknown>,
+  onAssistantText?: OneShotAgentParams["onAssistantText"],
 ): Promise<void> {
   const ts = new Date().toISOString().slice(11, 19);
   const type = typeof part.type === "string" ? part.type : "unknown";
 
   if (type === "text") {
     const text = typeof part.text === "string" ? part.text : "";
-    if (text) await appendLog(`\n## [${ts}] Assistant\n${text}\n`);
+    if (text) {
+      emitAssistantText(onAssistantText, text);
+      await appendLog(`\n## [${ts}] Assistant\n${text}\n`);
+    }
     return;
   }
 

package/src/backend/runtime/index.ts

@@ -20,7 +20,7 @@
  *   - `prompt/prompt-format` — user-prompt formatter
  *     ([time] [Name] [msg_id:N]).
  *   - `frontends` — config `frontend` → messaging-frontend list.
- *   - `extractSessionName` — re-exported from `util/session-name` so the
+ *   - `extractSessionName` — re-exported from `core/weaver/session-name` so the
  *     backends keep one import site; the helper itself is frontend-neutral.
  *   - `usage` — cache-hit % + log summarisers.
  *   - `prompt/system-prompt` — per-session prompt snapshots + backend
@@ -37,7 +37,7 @@
  *   - SDK-specific event types — those live in each backend.
  *   - Session storage — that's `src/storage/sessions.ts`.
  *   - MCP server registration — backend-specific transport details
- *     (the spawn/env contract they share is `core/tools/mcp-env.ts`).
+ *     (the spawn/env contract they share is `core/tools/ops/mcp-env.ts`).
  */
 
 export { captureDeliveredText } from "./turn/delivered-text.js";

package/src/backend/runtime/one-shot-hooks.ts

@@ -0,0 +1,45 @@
+/**
+ * One-shot run hooks — the shared, throw-proof call site for
+ * `OneShotAgentParams.onAssistantText`.
+ *
+ * Every backend's isolated runner (`BackgroundRunner.runOneShotAgent`) already
+ * renders the model's assistant text into the run log as markdown. The hook is
+ * the same text as data, so a caller (the sub-agent runner) can take a run's
+ * result without parsing the log back apart. It lives here rather than in each
+ * backend so all four report it identically: same "final answer only" meaning,
+ * same empty-string skip, same swallow-and-log on a throwing callback.
+ *
+ * Contract:
+ *   - Final assistant text only. Reasoning/thinking blocks and tool-call
+ *     payloads are log-only; they never reach the hook.
+ *   - Synchronous. The runner does not await the consumer, so a hook that
+ *     wants to do async work owns its own queueing.
+ *   - Never throws into the run. A consumer bug must not abort a heartbeat.
+ */
+
+import type { OneShotAgentParams } from "../../core/types.js";
+import { logWarn } from "../../util/log.js";
+
+/**
+ * Report one assistant text segment to the run's optional consumer.
+ *
+ * No-ops when there is no hook (the heartbeat/dream/cron path) or when the
+ * text is empty — the backends guard their log appends the same way, so the
+ * hook sees exactly the segments the log does.
+ */
+export function emitAssistantText(
+  onAssistantText: OneShotAgentParams["onAssistantText"],
+  text: string,
+): void {
+  if (!onAssistantText || !text) return;
+  try {
+    onAssistantText(text);
+  } catch (err) {
+    logWarn(
+      "agent",
+      `one-shot onAssistantText hook threw (ignored): ${
+        err instanceof Error ? err.message : String(err)
+      }`,
+    );
+  }
+}

package/src/backend/runtime/turn/delivery.ts

@@ -9,7 +9,7 @@
  * Routes:
  *
  *   - `tool` — a delivery tool (`end_turn` / `send` / `react`) already
- *     bridged the message to the platform via `core/tools/messaging.ts`.
+ *     bridged the message to the platform via `core/tools/chat/messaging.ts`.
  *     The message is already in the chat; we don't re-emit. We may
  *     have captured `deliveredTextNorms` for dedup, or just have
  *     `hadBridgeDelivery` set (e.g. `send(type="photo")` — non-text

package/src/backend/runtime/turn/turn-phases.ts

@@ -25,7 +25,7 @@ import {
   setSessionName,
 } from "../../../storage/sessions.js";
 import { log } from "../../../util/log.js";
-import { extractSessionName } from "../../../util/session-name.js";
+import { extractSessionName } from "../../../core/weaver/session-name.js";
 import { traceMessage } from "../../../util/trace.js";
 import {
   FLOW_VIOLATION_MAX_RETRIES,

package/src/bootstrap.ts

@@ -10,7 +10,7 @@
  */
 
 import { loadConfig, rebuildSystemPrompt } from "./core/config/index.js";
-import { initWorkspace } from "./util/workspace.js";
+import { initWorkspace } from "./core/vfs/workspace.js";
 import { loadSessions, resetSession } from "./storage/sessions.js";
 import { loadChatSettings } from "./storage/chat-settings.js";
 import { loadCronJobs } from "./storage/cron.js";
@@ -36,7 +36,7 @@ import {
 import { initDream, maybeStartDream } from "./core/background/dream/index.js";
 import { initHeartbeat } from "./core/background/heartbeat/index.js";
 import { log, logWarn, logDebug } from "./util/log.js";
-import { bootPhase } from "./util/boot-timer.js";
+import { bootPhase } from "./core/daemon/boot-timer.js";
 import { mapConcurrent } from "./util/concurrency.js";
 import type { TalonConfig } from "./core/config/index.js";
 import { resolveFrontendIdAmong } from "./core/frontend-runtime/routing.js";

package/src/cli.ts

@@ -16,13 +16,16 @@
  * `cli/index.ts`. Everything else lives under `cli/`.
  */
 
-import { MCP_LAUNCH_SUBCOMMAND, runSupervisor } from "./util/mcp-launcher.js";
+import {
+  MCP_LAUNCH_SUBCOMMAND,
+  runSupervisor,
+} from "./core/mcp-hub/launcher.js";
 import { LUA_RUN_SUBCOMMAND, runLuaMain } from "./core/scripts/lua.js";
 
 // Hidden subcommand dispatch — must run before anything else. Talon
 // supervises MCP stdio children (`_mcp-launch`) and runs WASM-sandboxed
 // Lua trigger scripts (`_lua-run`) by re-invoking its own entrypoint
-// (see util/mcp-launcher.ts). Neither call resolves; the helper process
+// (see core/mcp-hub/launcher.ts). Neither call resolves; the helper process
 // exits from its own handlers.
 if (process.argv[2] === MCP_LAUNCH_SUBCOMMAND) {
   await runSupervisor(process.argv.slice(3));

package/src/core/agent-runtime/README.md

@@ -69,6 +69,21 @@ onto the SQLite layer in `src/storage/` (typed tables or the `kv`
 singleton store). Legacy files import once at boot via
 `storage/legacy-import.ts` and are renamed `*.imported`.
 
+### `agent-host.ts`
+
+The seam the Claude Agent SDK moves out of the daemon across
+(`docs/agent-host-sidecar.md`). Two halves: the NDJSON wire vocabulary
+(`HostRequest` / `HostReply` / `HostEvent` / `HostNotice`, plus
+`parseHostMessage` / `serializeHostMessage` — unknown types parse to a
+typed `unknown` the reader logs and drops, never a throw), and
+`AgentHostClient`, the interface the daemon holds. Phase 1 ships one
+implementation, `backend/claude-sdk/host/in-process.ts`; Phase 2 adds a
+process-backed one behind the same interface. Fixtures:
+`protocol/fixtures/agent-host_v1.json`.
+
+The interface and codec live here, not in `backend/`, because `core/` may
+not import `backend/` — the implementations live under `backend/`.
+
 ### `contract-tests.ts`
 
 Backend contract assertions any conforming `Backend` must pass: