@pikiloom/kernel 0.3.3 → 0.3.5

package/dist/drivers/claude.d.ts

@@ -24,7 +24,7 @@ export declare function claudeBgAgentHoldCapMs(): number;
 export declare function claudeTurnHasAgentBackground(s: any): boolean;
 export declare function claudeBgHoldRecheckMs(): number;
 export declare function claudeBgSettleQuietMs(): number;
-export declare function claudeModelStallMs(): number;
+export declare function claudeModelStallMs(effort?: string | null): number;
 export declare const CLAUDE_TRUNCATED_RECOVERY_PROMPT: string;
 export declare function claudeTruncatedRecoveryEnabled(): boolean;
 export declare function isClaudeSyntheticResumeNoise(text: string): boolean;

package/dist/drivers/claude.js

@@ -193,7 +193,7 @@ export class ClaudeDriver {
                         return;
                     }
                     settleResult({ stopReason: 'stalled', kill: false, ok: false });
-                }, claudeModelStallMs());
+                }, claudeModelStallMs(input.effort));
                 unref(modelStallTimer);
             };
             try {
@@ -540,6 +540,23 @@ export function handleClaudeEvent(ev, s, emit) {
         // as the reply nor counts as real output (the no-op-resume recovery keys off that emptiness).
         if (ev.message?.model === '<synthetic>' && isClaudeSyntheticResumeNoise(claudeContentText(ev.message?.content)))
             return;
+        // API-error message: Claude surfaces a failed model call (401 auth, overloaded, quota, …) not as a
+        // result code but as a synthetic assistant message — model '<synthetic>', a lone text block carrying
+        // the human-readable error ("Failed to authenticate. API Error: 401 …"), and a TOP-LEVEL `error` tag
+        // on the event (e.g. "authentication_failed"; the persisted transcript also stamps `isApiErrorMessage`).
+        // It is NOT model output. Routing its text through the normal path below would make the error render
+        // as the assistant's reply body (原文). Send it to `s.error` — the run-end notice, same slot as the
+        // `result{is_error}` branch below — and never to `s.text`. The trailing `result` also flags the error
+        // (and would set `s.error` too), but claiming it here keeps a narration-less turn from double-rendering
+        // (body + notice), and preserves any real narration already streamed before the call failed.
+        const apiErrorTag = typeof ev.error === 'string' ? ev.error.trim() : (ev.isApiErrorMessage ? 'api_error' : '');
+        if (apiErrorTag) {
+            if (!s.error) {
+                const msg = claudeContentText(ev.message?.content).trim();
+                s.error = msg || `Claude reported an API error (${apiErrorTag})`;
+            }
+            return;
+        }
         const contents = ev.message?.content || [];
         for (const b of contents) {
             if (b?.type !== 'tool_use')
@@ -784,15 +801,22 @@ export function claudeBgSettleQuietMs() {
     return Number.isFinite(raw) && raw > 0 ? raw : CLAUDE_BG_SETTLE_QUIET_DEFAULT_MS;
 }
 // How long the model may stay COMPLETELY silent after a tool_result (control handed back to it,
-// no background pending) before the driver gives up and settles the turn as 'stalled'. Deliberately
-// generous: legitimate silent extended-thinking (subscription accounts stream no thinking text) and
-// slow providers must not trip it, and a still-running tool never trips it (it has no tool_result
-// yet). This is the backstop for a turn that would otherwise hang forever with the answer never
-// delivered. Override with PIKILOOM_CLAUDE_MODEL_STALL_MS.
-const CLAUDE_MODEL_STALL_DEFAULT_MS = 120_000;
-export function claudeModelStallMs() {
+// no background pending) before the driver gives up and settles the turn as 'stalled'. Must be
+// generous: subscription accounts stream NO events during extended thinking, and at the reasoning
+// rungs a legitimate silent think regularly exceeds two minutes — the original 120s default
+// misfired on exactly that (settling a LIVE turn as 'stalled' and then killing its still-running
+// tool via the leak-guard; mirasim#111). A too-long window only means a truly hung turn shows its
+// dead spinner longer, so the costs are asymmetric — err long. Effort-laddered: the deep-reasoning
+// rungs (high and up) think the longest. A still-running tool never trips this (it has no
+// tool_result yet). Override with PIKILOOM_CLAUDE_MODEL_STALL_MS (wins over the ladder).
+const CLAUDE_MODEL_STALL_DEFAULT_MS = 300_000;
+const CLAUDE_MODEL_STALL_DEEP_MS = 600_000;
+const CLAUDE_DEEP_REASONING_EFFORTS = new Set(['high', 'xhigh', 'max', 'ultra']);
+export function claudeModelStallMs(effort) {
     const raw = Number(process.env.PIKILOOM_CLAUDE_MODEL_STALL_MS);
-    return Number.isFinite(raw) && raw > 0 ? raw : CLAUDE_MODEL_STALL_DEFAULT_MS;
+    if (Number.isFinite(raw) && raw > 0)
+        return raw;
+    return effort && CLAUDE_DEEP_REASONING_EFFORTS.has(effort) ? CLAUDE_MODEL_STALL_DEEP_MS : CLAUDE_MODEL_STALL_DEFAULT_MS;
 }
 // In-process self-heal for a truncated turn: when a clean result lands while the tool loop is
 // still dangling (the model's closing round came back empty), the stdin is still open — inject

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pikiloom/kernel",
-  "version": "0.3.3",
+  "version": "0.3.5",
   "description": "Heterogeneous coding agents (Claude / Codex / Gemini / ACP) -> an interaction-friendly, accumulating session snapshot + control handle, over pluggable surfaces (IM, Web, tunnel, TUI). The reusable core pikiloom itself is built on.",
   "type": "module",
   "license": "MIT",