From 1c24dd28858c86dd123b3230d82fff87a114b9d3 Mon Sep 17 00:00:00 2001
From: Jiayuan Zhang <forrestchang7@gmail.com>
Date: Sun, 15 Feb 2026 17:48:21 +0800
Subject: [PATCH 01/12] fix(credentials): add fallback to ~/.super-multica for
 custom data dirs

When SMC_DATA_DIR is set (e.g., for E2E tests), the credentials lookup
now falls back to ~/.super-multica/credentials.json5 if the custom
data dir doesn't have its own credentials file. This mirrors the
existing fallback pattern in auth-store.ts and removes the need for
the SMC_CREDENTIALS_PATH workaround in E2E tests.

Lookup order:
1. SMC_CREDENTIALS_PATH env var (explicit override)
2. {DATA_DIR}/credentials.json5 (current data dir)
3. ~/.super-multica/credentials.json5 (default location fallback)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---
 packages/core/src/agent/credentials.ts | 30 ++++++++++++++++++++++++--
 1 file changed, 28 insertions(+), 2 deletions(-)

diff --git a/packages/core/src/agent/credentials.ts b/packages/core/src/agent/credentials.ts
index 6e6f8363..d93446b9 100644
--- a/packages/core/src/agent/credentials.ts
+++ b/packages/core/src/agent/credentials.ts
@@ -36,6 +36,7 @@ export type CredentialsConfig = {
 };
 
 const DEFAULT_CREDENTIALS_PATH = join(DATA_DIR, "credentials.json5");
+const FALLBACK_CREDENTIALS_PATH = join(homedir(), ".super-multica", "credentials.json5");
 
 function expandHome(value: string): string {
   if (value === "~") return homedir();
@@ -53,9 +54,34 @@ function isTestEnv(): boolean {
   );
 }
 
+/**
+ * Resolve the credentials file path.
+ *
+ * Lookup order:
+ * 1. SMC_CREDENTIALS_PATH env var (explicit override)
+ * 2. {DATA_DIR}/credentials.json5 (current data dir, respects SMC_DATA_DIR)
+ * 3. ~/.super-multica/credentials.json5 (default location fallback —
+ *    allows E2E tests and other custom SMC_DATA_DIR setups to
+ *    share the production credentials)
+ */
 export function getCredentialsPath(): string {
-  const raw = process.env.SMC_CREDENTIALS_PATH ?? DEFAULT_CREDENTIALS_PATH;
-  return expandHome(raw);
+  // Explicit env override — use as-is
+  if (process.env.SMC_CREDENTIALS_PATH) {
+    return expandHome(process.env.SMC_CREDENTIALS_PATH);
+  }
+
+  // Primary: current DATA_DIR
+  if (existsSync(DEFAULT_CREDENTIALS_PATH)) {
+    return DEFAULT_CREDENTIALS_PATH;
+  }
+
+  // Fallback: default ~/.super-multica location when using a custom data dir
+  if (DEFAULT_CREDENTIALS_PATH !== FALLBACK_CREDENTIALS_PATH && existsSync(FALLBACK_CREDENTIALS_PATH)) {
+    return FALLBACK_CREDENTIALS_PATH;
+  }
+
+  // Return primary path even if it doesn't exist (for error messages / creation)
+  return DEFAULT_CREDENTIALS_PATH;
 }
 
 export class CredentialManager {

From 755ed5e9dee6bd9fbd57a3c9b79594954bf24915 Mon Sep 17 00:00:00 2001
From: Jiayuan Zhang <forrestchang7@gmail.com>
Date: Sun, 15 Feb 2026 18:06:42 +0800
Subject: [PATCH 02/12] feat(run-log): add result metadata to tool_end events

Enrich tool_end events with result_chars, result_summary, and
error_type fields. Since run-log.jsonl is append-only and never
compacted, this preserves tool result metadata that would otherwise
be lost when session.jsonl undergoes compaction.

New fields:
- result_chars: total character count of result content
- result_summary: short tool-specific summary (e.g. "10 results",
  "12.5KB", "finance/get_price_snapshot")
- error_type: error category for tool errors (e.g. "fetch_failed")

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---
 packages/core/src/agent/run-log.ts |  5 ++-
 packages/core/src/agent/runner.ts  | 62 +++++++++++++++++++++++++++++-
 2 files changed, 64 insertions(+), 3 deletions(-)

diff --git a/packages/core/src/agent/run-log.ts b/packages/core/src/agent/run-log.ts
index e8e3b446..18bf5c80 100644
--- a/packages/core/src/agent/run-log.ts
+++ b/packages/core/src/agent/run-log.ts
@@ -25,7 +25,10 @@
  * - `tool_start`  — Tool execution begins.
  *     Fields: tool (name), args (first 500 chars of JSON)
  * - `tool_end`    — Tool execution completes.
- *     Fields: tool (name), duration_ms, is_error
+ *     Fields: tool (name), duration_ms, is_error, result_chars, result_summary?, error_type?
+ *     result_chars: total character count of result content (survives session compaction)
+ *     result_summary: short tool-specific summary (e.g. "10 results", "12.5KB", "finance/get_price_snapshot")
+ *     error_type: error category when tool returned an error (e.g. "fetch_failed", "ssrf_blocked")
  *
  * ### Context Management — Preflight (before LLM call)
  * - `preflight_compact_start` — Preflight compaction triggered.
diff --git a/packages/core/src/agent/runner.ts b/packages/core/src/agent/runner.ts
index e561213d..ba652939 100644
--- a/packages/core/src/agent/runner.ts
+++ b/packages/core/src/agent/runner.ts
@@ -85,6 +85,51 @@ export function isRotatableError(reason: AuthProfileFailureReason): boolean {
   return reason === "auth" || reason === "rate_limit" || reason === "billing" || reason === "timeout";
 }
 
+// ── Run-log result extraction helpers ──────────────────────────────────────
+// Lightweight extractors for tool_end metadata. These mirror the patterns in
+// cli/output.ts but are kept separate to avoid CLI-specific dependencies.
+
+function extractRunLogResultText(result: unknown): string | undefined {
+  if (!result || typeof result !== "object") return undefined;
+  const msg = result as { content?: Array<{ type: string; text?: string }> };
+  if (Array.isArray(msg.content)) {
+    for (const c of msg.content) {
+      if (c.type === "text" && c.text) return c.text;
+    }
+  }
+  return undefined;
+}
+
+function extractRunLogResultDetails(result: unknown): Record<string, unknown> | null {
+  const text = extractRunLogResultText(result);
+  if (text) {
+    try { return JSON.parse(text) as Record<string, unknown>; } catch { /* non-JSON result */ }
+  }
+  const withDetails = result as { details?: unknown };
+  if (withDetails?.details && typeof withDetails.details === "object") {
+    return withDetails.details as Record<string, unknown>;
+  }
+  return null;
+}
+
+function formatRunLogToolSummary(tool: string, details: Record<string, unknown> | null): string | undefined {
+  if (!details) return undefined;
+  if (details.error) return `error: ${details.message || details.error}`;
+  switch (tool) {
+    case "web_search": return `${details.count ?? 0} results`;
+    case "web_fetch": {
+      const parts: string[] = [];
+      if (typeof details.length === "number") parts.push(`${(details.length as number / 1024).toFixed(1)}KB`);
+      if (details.cached) parts.push("cached");
+      return parts.join(", ") || undefined;
+    }
+    case "data": return `${details.domain}/${details.action}`;
+    case "glob": return `${details.count ?? 0} files`;
+    case "exec": return details.exitCode !== undefined ? `exit ${details.exitCode}` : undefined;
+    default: return undefined;
+  }
+}
+
 export class Agent {
   private readonly agent: PiAgentCore;
   private output;
@@ -746,11 +791,24 @@ export class Agent {
       const startTime = this.toolStartTimes.get(toolName);
       const duration_ms = startTime ? Date.now() - startTime : undefined;
       this.toolStartTimes.delete(toolName);
-      this.runLog.log("tool_end", {
+
+      // Extract result metadata for run-log persistence (survives session compaction)
+      const result = (event as any).result;
+      const resultText = extractRunLogResultText(result);
+      const resultChars = resultText?.length ?? 0;
+      const details = extractRunLogResultDetails(result);
+
+      const toolEndData: Record<string, unknown> = {
         tool: toolName,
         duration_ms,
         is_error: (event as any).isError ?? false,
-      });
+        result_chars: resultChars,
+        result_summary: formatRunLogToolSummary(toolName, details),
+      };
+      if (details?.error) {
+        toolEndData.error_type = String(details.error);
+      }
+      this.runLog.log("tool_end", toolEndData);
     }
   }
 

From b007ddffc8b8f9a162be3be888a1ef68489808d9 Mon Sep 17 00:00:00 2001
From: Jiayuan Zhang <forrestchang7@gmail.com>
Date: Sun, 15 Feb 2026 18:06:51 +0800
Subject: [PATCH 03/12] feat(cli): initialize Hub in run mode for full agent
 capabilities
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Always initialize Hub in CLI run mode to match the Desktop environment
where Hub is always active. This enables sessions_spawn (sub-agent
creation), cron tasks, channel plugins, and other Hub-dependent
features during E2E testing.

Hub constructor is non-blocking — gateway connection failures are
handled gracefully with auto-reconnect. hub.shutdown() in finally
block ensures clean teardown on exit.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---
 apps/cli/src/commands/run.ts | 66 +++++++++++++++++++++---------------
 1 file changed, 38 insertions(+), 28 deletions(-)

diff --git a/apps/cli/src/commands/run.ts b/apps/cli/src/commands/run.ts
index 1b915d0d..c5fb6916 100644
--- a/apps/cli/src/commands/run.ts
+++ b/apps/cli/src/commands/run.ts
@@ -7,7 +7,7 @@
  */
 
 import { join } from "node:path";
-import { Agent } from "@multica/core";
+import { Agent, Hub } from "@multica/core";
 import type { AgentOptions } from "@multica/core";
 import type { ToolsConfig } from "@multica/core";
 import { DATA_DIR } from "@multica/utils";
@@ -192,35 +192,45 @@ export async function runCommand(args: string[]): Promise<void> {
 
   const enableRunLog = opts.runLog || !!process.env.MULTICA_RUN_LOG;
 
-  const agent = new Agent({
-    profileId: opts.profile,
-    provider: opts.provider,
-    model: opts.model,
-    apiKey: opts.apiKey,
-    baseUrl: opts.baseUrl,
-    systemPrompt: opts.system,
-    thinkingLevel: opts.thinking as any,
-    reasoningMode: opts.reasoning as AgentOptions["reasoningMode"],
-    cwd: opts.cwd,
-    sessionId: opts.session,
-    debug: opts.debug,
-    enableRunLog,
-    tools: toolsConfig,
-  });
+  // Initialize Hub to enable full agent capabilities (sub-agents, channels, cron).
+  // Matches Desktop environment where Hub is always active.
+  // Gateway connection failures are non-blocking (auto-reconnect with backoff).
+  const gatewayUrl = process.env.GATEWAY_URL || "http://localhost:3000";
+  const hub = new Hub(gatewayUrl);
 
-  const sessionDir = join(DATA_DIR, "sessions", agent.sessionId);
+  try {
+    const agent = new Agent({
+      profileId: opts.profile,
+      provider: opts.provider,
+      model: opts.model,
+      apiKey: opts.apiKey,
+      baseUrl: opts.baseUrl,
+      systemPrompt: opts.system,
+      thinkingLevel: opts.thinking as any,
+      reasoningMode: opts.reasoning as AgentOptions["reasoningMode"],
+      cwd: opts.cwd,
+      sessionId: opts.session,
+      debug: opts.debug,
+      enableRunLog,
+      tools: toolsConfig,
+    });
 
-  // If it's a newly created session, notify user of sessionId
-  if (!opts.session) {
-    console.error(`[session: ${agent.sessionId}]`);
-  }
-  if (enableRunLog) {
-    console.error(`[session-dir: ${sessionDir}]`);
-  }
+    const sessionDir = join(DATA_DIR, "sessions", agent.sessionId);
 
-  const result = await agent.run(finalPrompt);
-  if (result.error) {
-    console.error(`Error: ${result.error}`);
-    process.exitCode = 1;
+    // If it's a newly created session, notify user of sessionId
+    if (!opts.session) {
+      console.error(`[session: ${agent.sessionId}]`);
+    }
+    if (enableRunLog) {
+      console.error(`[session-dir: ${sessionDir}]`);
+    }
+
+    const result = await agent.run(finalPrompt);
+    if (result.error) {
+      console.error(`Error: ${result.error}`);
+      process.exitCode = 1;
+    }
+  } finally {
+    hub.shutdown();
   }
 }

From 02ed09b77b15bddbec5c49d6d9e7e22d0fd48743 Mon Sep 17 00:00:00 2001
From: Jiayuan Zhang <forrestchang7@gmail.com>
Date: Sun, 15 Feb 2026 18:47:37 +0800
Subject: [PATCH 04/12] fix(tools): use boolean error flag in web_fetch and
 web_search error responses

Return error: true (boolean) with code field instead of error: "string_code"
to match ToolErrorPayload convention. Also update runner.ts formatRunLogToolSummary
to prefer details.code for error categorization.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---
 packages/core/src/agent/tools/web/web-fetch.ts  | 6 ++++--
 packages/core/src/agent/tools/web/web-search.ts | 3 ++-
 2 files changed, 6 insertions(+), 3 deletions(-)

diff --git a/packages/core/src/agent/tools/web/web-fetch.ts b/packages/core/src/agent/tools/web/web-fetch.ts
index 32a1e079..476e9400 100644
--- a/packages/core/src/agent/tools/web/web-fetch.ts
+++ b/packages/core/src/agent/tools/web/web-fetch.ts
@@ -343,12 +343,14 @@ export function createWebFetchTool(): AgentTool<typeof WebFetchSchema, unknown>
       } catch (error) {
         if (error instanceof SsrfBlockedError) {
           return jsonResult({
-            error: "ssrf_blocked",
+            error: true,
+            code: "ssrf_blocked",
             message: error.message,
           });
         }
         return jsonResult({
-          error: "fetch_failed",
+          error: true,
+          code: "fetch_failed",
           message: error instanceof Error ? error.message : String(error),
         });
       }
diff --git a/packages/core/src/agent/tools/web/web-search.ts b/packages/core/src/agent/tools/web/web-search.ts
index 5d2bece1..5dff6ed8 100644
--- a/packages/core/src/agent/tools/web/web-search.ts
+++ b/packages/core/src/agent/tools/web/web-search.ts
@@ -135,7 +135,8 @@ export function createWebSearchTool(): AgentTool<typeof WebSearchSchema, unknown
         return jsonResult(result);
       } catch (error) {
         return jsonResult({
-          error: "search_failed",
+          error: true,
+          code: "search_failed",
           message: error instanceof Error ? error.message : String(error),
         });
       }

From c012bff246350d4ed384802a3369bdffe41cc45e Mon Sep 17 00:00:00 2001
From: Jiayuan Zhang <forrestchang7@gmail.com>
Date: Sun, 15 Feb 2026 18:47:44 +0800
Subject: [PATCH 05/12] fix(tools): show findings and full runId in
 sessions_list list view

Grouped runs now display findings for completed sub-agents (up to 4000
chars). Ungrouped runs increased truncation from 200 to 4000 chars. All
status lines include full runId for subsequent API queries.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---
 .../src/agent/tools/sessions-list.test.ts     | 49 +++++++++++++++++++
 .../core/src/agent/tools/sessions-list.ts     | 13 +++--
 2 files changed, 57 insertions(+), 5 deletions(-)

diff --git a/packages/core/src/agent/tools/sessions-list.test.ts b/packages/core/src/agent/tools/sessions-list.test.ts
index 230d8997..637a8537 100644
--- a/packages/core/src/agent/tools/sessions-list.test.ts
+++ b/packages/core/src/agent/tools/sessions-list.test.ts
@@ -47,6 +47,8 @@ describe("sessions_list tool", () => {
         startedAt: now - 60000,
         endedAt: now - 30000,
         outcome: { status: "ok" },
+        findings: "All tests passed successfully.",
+        findingsCaptured: true,
       }),
     );
     seedSubagentRunForTests(
@@ -56,6 +58,8 @@ describe("sessions_list tool", () => {
         startedAt: now - 60000,
         endedAt: now,
         outcome: { status: "error", error: "timeout" },
+        findings: "Lint check timed out.",
+        findingsCaptured: true,
       }),
     );
 
@@ -71,6 +75,13 @@ describe("sessions_list tool", () => {
     expect((text as { text: string }).text).toContain("Code Review");
     expect((text as { text: string }).text).toContain("Test Analysis");
     expect((text as { text: string }).text).toContain("Lint Check");
+    // Verify full runId is shown for completed runs
+    expect((text as { text: string }).text).toContain("id:run-aaa");
+    expect((text as { text: string }).text).toContain("id:run-bbb");
+    expect((text as { text: string }).text).toContain("id:run-ccc");
+    // Verify findings are shown for completed runs
+    expect((text as { text: string }).text).toContain("All tests passed successfully.");
+    expect((text as { text: string }).text).toContain("Lint check timed out.");
 
     expect(result.details!.runs).toHaveLength(3);
     expect(result.details!.runs[0]!.status).toBe("running");
@@ -141,6 +152,44 @@ describe("sessions_list tool", () => {
     expect(result.details).toEqual({ runs: [] });
   });
 
+  it("shows findings for grouped completed runs", async () => {
+    const now = Date.now();
+    const groupId = "group-001";
+    seedSubagentRunForTests(
+      makeRecord({
+        runId: "run-g1",
+        label: "Bull Case Research",
+        startedAt: now - 60000,
+        endedAt: now - 10000,
+        outcome: { status: "ok" },
+        findings: "AI infrastructure capex growing 40% YoY.",
+        findingsCaptured: true,
+        groupId,
+      }),
+    );
+    seedSubagentRunForTests(
+      makeRecord({
+        runId: "run-g2",
+        label: "Bear Case Research",
+        startedAt: now - 60000,
+        endedAt: now - 5000,
+        outcome: { status: "ok" },
+        findings: "Valuation risk: forward P/E above historical average.",
+        findingsCaptured: true,
+        groupId,
+      }),
+    );
+
+    const tool = createSessionsListTool({ sessionId: "parent-001" });
+    const result = await tool.execute("call-1", {});
+
+    const text = (result.content[0] as { text: string }).text;
+    expect(text).toContain("id:run-g1");
+    expect(text).toContain("id:run-g2");
+    expect(text).toContain("AI infrastructure capex growing 40% YoY.");
+    expect(text).toContain("Valuation risk: forward P/E above historical average.");
+  });
+
   it("shows findings status for running task", async () => {
     const now = Date.now();
     seedSubagentRunForTests(
diff --git a/packages/core/src/agent/tools/sessions-list.ts b/packages/core/src/agent/tools/sessions-list.ts
index 93ef896c..8905b943 100644
--- a/packages/core/src/agent/tools/sessions-list.ts
+++ b/packages/core/src/agent/tools/sessions-list.ts
@@ -212,10 +212,13 @@ export function createSessionsListTool(
           const status = resolveStatus(r);
           if (status === "running") {
             const elapsed = r.startedAt ? formatElapsed(now - r.startedAt) : "just spawned";
-            statusLines.push(`     ${idx}. [RUNNING] "${displayName}" (${elapsed})`);
+            statusLines.push(`     ${idx}. [RUNNING] "${displayName}" (${elapsed})  id:${r.runId}`);
           } else {
             const elapsed = r.startedAt && r.endedAt ? formatElapsed(r.endedAt - r.startedAt) : "";
-            statusLines.push(`     ${idx}. [${status.toUpperCase()}] "${displayName}" (${elapsed})`);
+            const findings = r.findingsCaptured
+              ? (r.findings ? r.findings.slice(0, 4000) + (r.findings.length > 4000 ? "…" : "") : "(no output)")
+              : "(findings not yet captured)";
+            statusLines.push(`     ${idx}. [${status.toUpperCase()}] "${displayName}" (${elapsed})  id:${r.runId}\n          Findings: ${findings}`);
           }
         }
       }
@@ -227,13 +230,13 @@ export function createSessionsListTool(
         const status = resolveStatus(r);
         if (status === "running") {
           const elapsed = r.startedAt ? formatElapsed(now - r.startedAt) : "just spawned";
-          statusLines.push(`  ${idx}. [RUNNING] "${displayName}" (${elapsed})`);
+          statusLines.push(`  ${idx}. [RUNNING] "${displayName}" (${elapsed})  id:${r.runId}`);
         } else {
           const elapsed = r.startedAt && r.endedAt ? formatElapsed(r.endedAt - r.startedAt) : "";
           const findings = r.findingsCaptured
-            ? (r.findings ? r.findings.slice(0, 200) + (r.findings.length > 200 ? "…" : "") : "(no output)")
+            ? (r.findings ? r.findings.slice(0, 4000) + (r.findings.length > 4000 ? "…" : "") : "(no output)")
             : "(findings not yet captured)";
-          statusLines.push(`  ${idx}. [${status.toUpperCase()}] "${displayName}" (${elapsed})\n      Findings: ${findings}`);
+          statusLines.push(`  ${idx}. [${status.toUpperCase()}] "${displayName}" (${elapsed})  id:${r.runId}\n      Findings: ${findings}`);
         }
       }
 

From a254daff013ce39f0d185d9057329fab0774302f Mon Sep 17 00:00:00 2001
From: Jiayuan Zhang <forrestchang7@gmail.com>
Date: Sun, 15 Feb 2026 18:47:53 +0800
Subject: [PATCH 06/12] feat(agent): enable parallel tool execution via
 pi-agent-core patch

Replace sequential for+await tool dispatch with Promise.allSettled for
parallel execution. All tool_execution_start events emit immediately,
tools run concurrently, results are processed in original order.

Also fix run-log toolStartTimes to key by toolCallId instead of toolName
to prevent collisions with parallel same-name tools.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---
 package.json                                  |  3 +
 packages/core/src/agent/runner.ts             | 12 +--
 .../@mariozechner__pi-agent-core@0.52.9.patch | 80 +++++++++++++++++++
 pnpm-lock.yaml                                | 35 ++++----
 4 files changed, 110 insertions(+), 20 deletions(-)
 create mode 100644 patches/@mariozechner__pi-agent-core@0.52.9.patch

diff --git a/package.json b/package.json
index 80b868f4..d39e6145 100644
--- a/package.json
+++ b/package.json
@@ -44,6 +44,9 @@
     "overrides": {
       "@types/react": "catalog:",
       "@types/react-dom": "catalog:"
+    },
+    "patchedDependencies": {
+      "@mariozechner/pi-agent-core@0.52.9": "patches/@mariozechner__pi-agent-core@0.52.9.patch"
     }
   },
   "devDependencies": {
diff --git a/packages/core/src/agent/runner.ts b/packages/core/src/agent/runner.ts
index ba652939..c9f81f03 100644
--- a/packages/core/src/agent/runner.ts
+++ b/packages/core/src/agent/runner.ts
@@ -114,7 +114,7 @@ function extractRunLogResultDetails(result: unknown): Record<string, unknown> |
 
 function formatRunLogToolSummary(tool: string, details: Record<string, unknown> | null): string | undefined {
   if (!details) return undefined;
-  if (details.error) return `error: ${details.message || details.error}`;
+  if (details.error) return `error: ${details.code || details.message || details.error}`;
   switch (tool) {
     case "web_search": return `${details.count ?? 0} results`;
     case "web_fetch": {
@@ -780,17 +780,19 @@ export class Agent {
 
   private handleRunLogEvent(event: AgentEvent) {
     if (event.type === "tool_execution_start") {
+      const toolCallId = (event as any).toolCallId ?? "unknown";
       const toolName = (event as any).toolName ?? "unknown";
-      this.toolStartTimes.set(toolName, Date.now());
+      this.toolStartTimes.set(toolCallId, Date.now());
       this.runLog.log("tool_start", {
         tool: toolName,
         args: JSON.stringify((event as any).args ?? {}).slice(0, 500),
       });
     } else if (event.type === "tool_execution_end") {
+      const toolCallId = (event as any).toolCallId ?? "unknown";
       const toolName = (event as any).toolName ?? "unknown";
-      const startTime = this.toolStartTimes.get(toolName);
+      const startTime = this.toolStartTimes.get(toolCallId);
       const duration_ms = startTime ? Date.now() - startTime : undefined;
-      this.toolStartTimes.delete(toolName);
+      this.toolStartTimes.delete(toolCallId);
 
       // Extract result metadata for run-log persistence (survives session compaction)
       const result = (event as any).result;
@@ -806,7 +808,7 @@ export class Agent {
         result_summary: formatRunLogToolSummary(toolName, details),
       };
       if (details?.error) {
-        toolEndData.error_type = String(details.error);
+        toolEndData.error_type = details.code ? String(details.code) : String(details.error);
       }
       this.runLog.log("tool_end", toolEndData);
     }
diff --git a/patches/@mariozechner__pi-agent-core@0.52.9.patch b/patches/@mariozechner__pi-agent-core@0.52.9.patch
new file mode 100644
index 00000000..e570a653
--- /dev/null
+++ b/patches/@mariozechner__pi-agent-core@0.52.9.patch
@@ -0,0 +1,80 @@
+diff --git a/dist/agent-loop.js b/dist/agent-loop.js
+index b26d45753809da14df6409354e8c537684b9acd7..931399cc12bafea940fe99ab74ae288d71506e52 100644
+--- a/dist/agent-loop.js
++++ b/dist/agent-loop.js
+@@ -206,17 +206,18 @@ async function streamAssistantResponse(context, config, signal, stream, streamFn
+  */
+ async function executeToolCalls(tools, assistantMessage, signal, stream, getSteeringMessages) {
+     const toolCalls = assistantMessage.content.filter((c) => c.type === "toolCall");
+-    const results = [];
+-    let steeringMessages;
+-    for (let index = 0; index < toolCalls.length; index++) {
+-        const toolCall = toolCalls[index];
+-        const tool = tools?.find((t) => t.name === toolCall.name);
++    // Emit all tool_execution_start events immediately
++    for (const toolCall of toolCalls) {
+         stream.push({
+             type: "tool_execution_start",
+             toolCallId: toolCall.id,
+             toolName: toolCall.name,
+             args: toolCall.arguments,
+         });
++    }
++    // Execute all tools in parallel
++    const settled = await Promise.allSettled(toolCalls.map(async (toolCall) => {
++        const tool = tools?.find((t) => t.name === toolCall.name);
+         let result;
+         let isError = false;
+         try {
+@@ -240,6 +241,27 @@ async function executeToolCalls(tools, assistantMessage, signal, stream, getStee
+             };
+             isError = true;
+         }
++        return { result, isError };
++    }));
++    // Process results IN ORIGINAL ORDER (critical for LLM context)
++    const results = [];
++    let steeringMessages;
++    for (let i = 0; i < settled.length; i++) {
++        const entry = settled[i];
++        const toolCall = toolCalls[i];
++        let result;
++        let isError;
++        if (entry.status === "fulfilled") {
++            result = entry.value.result;
++            isError = entry.value.isError;
++        }
++        else {
++            result = {
++                content: [{ type: "text", text: entry.reason instanceof Error ? entry.reason.message : String(entry.reason) }],
++                details: {},
++            };
++            isError = true;
++        }
+         stream.push({
+             type: "tool_execution_end",
+             toolCallId: toolCall.id,
+@@ -259,17 +281,12 @@ async function executeToolCalls(tools, assistantMessage, signal, stream, getStee
+         results.push(toolResultMessage);
+         stream.push({ type: "message_start", message: toolResultMessage });
+         stream.push({ type: "message_end", message: toolResultMessage });
+-        // Check for steering messages - skip remaining tools if user interrupted
+-        if (getSteeringMessages) {
+-            const steering = await getSteeringMessages();
+-            if (steering.length > 0) {
+-                steeringMessages = steering;
+-                const remainingCalls = toolCalls.slice(index + 1);
+-                for (const skipped of remainingCalls) {
+-                    results.push(skipToolCall(skipped, stream));
+-                }
+-                break;
+-            }
++    }
++    // Check steering messages once after all tools complete
++    if (getSteeringMessages) {
++        const steering = await getSteeringMessages();
++        if (steering.length > 0) {
++            steeringMessages = steering;
+         }
+     }
+     return { toolResults: results, steeringMessages };
diff --git a/pnpm-lock.yaml b/pnpm-lock.yaml
index a0221107..18c13e8b 100644
--- a/pnpm-lock.yaml
+++ b/pnpm-lock.yaml
@@ -68,13 +68,18 @@ overrides:
   '@types/react': ^19.2.0
   '@types/react-dom': ^19.2.0
 
+patchedDependencies:
+  '@mariozechner/pi-agent-core@0.52.9':
+    hash: befbe3b9fd1a6d3e13dfa4927d9f0addd2f3b3454bd9d407c16dacb9a71c0f00
+    path: patches/@mariozechner__pi-agent-core@0.52.9.patch
+
 importers:
 
   .:
     dependencies:
       '@mariozechner/pi-agent-core':
         specifier: 'catalog:'
-        version: 0.52.9(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
+        version: 0.52.9(patch_hash=befbe3b9fd1a6d3e13dfa4927d9f0addd2f3b3454bd9d407c16dacb9a71c0f00)(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
       '@mariozechner/pi-ai':
         specifier: 'catalog:'
         version: 0.52.9(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
@@ -591,7 +596,7 @@ importers:
     dependencies:
       '@mariozechner/pi-agent-core':
         specifier: 'catalog:'
-        version: 0.52.9(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
+        version: 0.52.9(patch_hash=befbe3b9fd1a6d3e13dfa4927d9f0addd2f3b3454bd9d407c16dacb9a71c0f00)(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
       '@mariozechner/pi-ai':
         specifier: 'catalog:'
         version: 0.52.9(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
@@ -690,7 +695,7 @@ importers:
     devDependencies:
       '@mariozechner/pi-agent-core':
         specifier: 'catalog:'
-        version: 0.52.9(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
+        version: 0.52.9(patch_hash=befbe3b9fd1a6d3e13dfa4927d9f0addd2f3b3454bd9d407c16dacb9a71c0f00)(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
       '@mariozechner/pi-ai':
         specifier: 'catalog:'
         version: 0.52.9(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
@@ -12498,7 +12503,7 @@ snapshots:
       std-env: 3.10.0
       yoctocolors: 2.1.2
 
-  '@mariozechner/pi-agent-core@0.52.9(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)':
+  '@mariozechner/pi-agent-core@0.52.9(patch_hash=befbe3b9fd1a6d3e13dfa4927d9f0addd2f3b3454bd9d407c16dacb9a71c0f00)(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)':
     dependencies:
       '@mariozechner/pi-ai': 0.52.9(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
     transitivePeerDependencies:
@@ -12537,7 +12542,7 @@ snapshots:
   '@mariozechner/pi-coding-agent@0.52.9(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)':
     dependencies:
       '@mariozechner/jiti': 2.6.5
-      '@mariozechner/pi-agent-core': 0.52.9(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
+      '@mariozechner/pi-agent-core': 0.52.9(patch_hash=befbe3b9fd1a6d3e13dfa4927d9f0addd2f3b3454bd9d407c16dacb9a71c0f00)(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
       '@mariozechner/pi-ai': 0.52.9(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
       '@mariozechner/pi-tui': 0.52.9
       '@silvia-odwyer/photon-node': 0.3.4
@@ -15944,9 +15949,9 @@ snapshots:
       '@typescript-eslint/eslint-plugin': 8.55.0(@typescript-eslint/parser@8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3))(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3)
       '@typescript-eslint/parser': 8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3)
       eslint: 9.39.2(jiti@2.6.1)
-      eslint-import-resolver-typescript: 3.10.1(eslint-plugin-import@2.32.0)(eslint@9.39.2(jiti@2.6.1))
+      eslint-import-resolver-typescript: 3.10.1(eslint-plugin-import@2.32.0(@typescript-eslint/parser@8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3))(eslint@9.39.2(jiti@2.6.1)))(eslint@9.39.2(jiti@2.6.1))
       eslint-plugin-expo: 1.0.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3)
-      eslint-plugin-import: 2.32.0(@typescript-eslint/parser@8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3))(eslint-import-resolver-typescript@3.10.1)(eslint@9.39.2(jiti@2.6.1))
+      eslint-plugin-import: 2.32.0(@typescript-eslint/parser@8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3))(eslint-import-resolver-typescript@3.10.1(eslint-plugin-import@2.32.0(@typescript-eslint/parser@8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3))(eslint@9.39.2(jiti@2.6.1)))(eslint@9.39.2(jiti@2.6.1)))(eslint@9.39.2(jiti@2.6.1))
       eslint-plugin-react: 7.37.5(eslint@9.39.2(jiti@2.6.1))
       eslint-plugin-react-hooks: 5.2.0(eslint@9.39.2(jiti@2.6.1))
       globals: 16.5.0
@@ -15961,8 +15966,8 @@ snapshots:
       '@next/eslint-plugin-next': 16.1.6
       eslint: 9.39.2(jiti@2.6.1)
       eslint-import-resolver-node: 0.3.9
-      eslint-import-resolver-typescript: 3.10.1(eslint-plugin-import@2.32.0)(eslint@9.39.2(jiti@2.6.1))
-      eslint-plugin-import: 2.32.0(@typescript-eslint/parser@8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3))(eslint-import-resolver-typescript@3.10.1)(eslint@9.39.2(jiti@2.6.1))
+      eslint-import-resolver-typescript: 3.10.1(eslint-plugin-import@2.32.0(@typescript-eslint/parser@8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3))(eslint@9.39.2(jiti@2.6.1)))(eslint@9.39.2(jiti@2.6.1))
+      eslint-plugin-import: 2.32.0(@typescript-eslint/parser@8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3))(eslint-import-resolver-typescript@3.10.1(eslint-plugin-import@2.32.0(@typescript-eslint/parser@8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3))(eslint@9.39.2(jiti@2.6.1)))(eslint@9.39.2(jiti@2.6.1)))(eslint@9.39.2(jiti@2.6.1))
       eslint-plugin-jsx-a11y: 6.10.2(eslint@9.39.2(jiti@2.6.1))
       eslint-plugin-react: 7.37.5(eslint@9.39.2(jiti@2.6.1))
       eslint-plugin-react-hooks: 7.0.1(eslint@9.39.2(jiti@2.6.1))
@@ -15984,7 +15989,7 @@ snapshots:
     transitivePeerDependencies:
       - supports-color
 
-  eslint-import-resolver-typescript@3.10.1(eslint-plugin-import@2.32.0)(eslint@9.39.2(jiti@2.6.1)):
+  eslint-import-resolver-typescript@3.10.1(eslint-plugin-import@2.32.0(@typescript-eslint/parser@8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3))(eslint@9.39.2(jiti@2.6.1)))(eslint@9.39.2(jiti@2.6.1)):
     dependencies:
       '@nolyfill/is-core-module': 1.0.39
       debug: 4.4.3
@@ -15995,18 +16000,18 @@ snapshots:
       tinyglobby: 0.2.15
       unrs-resolver: 1.11.1
     optionalDependencies:
-      eslint-plugin-import: 2.32.0(@typescript-eslint/parser@8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3))(eslint-import-resolver-typescript@3.10.1)(eslint@9.39.2(jiti@2.6.1))
+      eslint-plugin-import: 2.32.0(@typescript-eslint/parser@8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3))(eslint-import-resolver-typescript@3.10.1(eslint-plugin-import@2.32.0(@typescript-eslint/parser@8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3))(eslint@9.39.2(jiti@2.6.1)))(eslint@9.39.2(jiti@2.6.1)))(eslint@9.39.2(jiti@2.6.1))
     transitivePeerDependencies:
       - supports-color
 
-  eslint-module-utils@2.12.1(@typescript-eslint/parser@8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3))(eslint-import-resolver-node@0.3.9)(eslint-import-resolver-typescript@3.10.1)(eslint@9.39.2(jiti@2.6.1)):
+  eslint-module-utils@2.12.1(@typescript-eslint/parser@8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3))(eslint-import-resolver-node@0.3.9)(eslint-import-resolver-typescript@3.10.1(eslint-plugin-import@2.32.0(@typescript-eslint/parser@8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3))(eslint@9.39.2(jiti@2.6.1)))(eslint@9.39.2(jiti@2.6.1)))(eslint@9.39.2(jiti@2.6.1)):
     dependencies:
       debug: 3.2.7
     optionalDependencies:
       '@typescript-eslint/parser': 8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3)
       eslint: 9.39.2(jiti@2.6.1)
       eslint-import-resolver-node: 0.3.9
-      eslint-import-resolver-typescript: 3.10.1(eslint-plugin-import@2.32.0)(eslint@9.39.2(jiti@2.6.1))
+      eslint-import-resolver-typescript: 3.10.1(eslint-plugin-import@2.32.0(@typescript-eslint/parser@8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3))(eslint@9.39.2(jiti@2.6.1)))(eslint@9.39.2(jiti@2.6.1))
     transitivePeerDependencies:
       - supports-color
 
@@ -16019,7 +16024,7 @@ snapshots:
       - supports-color
       - typescript
 
-  eslint-plugin-import@2.32.0(@typescript-eslint/parser@8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3))(eslint-import-resolver-typescript@3.10.1)(eslint@9.39.2(jiti@2.6.1)):
+  eslint-plugin-import@2.32.0(@typescript-eslint/parser@8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3))(eslint-import-resolver-typescript@3.10.1(eslint-plugin-import@2.32.0(@typescript-eslint/parser@8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3))(eslint@9.39.2(jiti@2.6.1)))(eslint@9.39.2(jiti@2.6.1)))(eslint@9.39.2(jiti@2.6.1)):
     dependencies:
       '@rtsao/scc': 1.1.0
       array-includes: 3.1.9
@@ -16030,7 +16035,7 @@ snapshots:
       doctrine: 2.1.0
       eslint: 9.39.2(jiti@2.6.1)
       eslint-import-resolver-node: 0.3.9
-      eslint-module-utils: 2.12.1(@typescript-eslint/parser@8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3))(eslint-import-resolver-node@0.3.9)(eslint-import-resolver-typescript@3.10.1)(eslint@9.39.2(jiti@2.6.1))
+      eslint-module-utils: 2.12.1(@typescript-eslint/parser@8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3))(eslint-import-resolver-node@0.3.9)(eslint-import-resolver-typescript@3.10.1(eslint-plugin-import@2.32.0(@typescript-eslint/parser@8.55.0(eslint@9.39.2(jiti@2.6.1))(typescript@5.9.3))(eslint@9.39.2(jiti@2.6.1)))(eslint@9.39.2(jiti@2.6.1)))(eslint@9.39.2(jiti@2.6.1))
       hasown: 2.0.2
       is-core-module: 2.16.1
       is-glob: 4.0.3

From d162ba98a9b6894ea8f3f24e49015e4c31d795af Mon Sep 17 00:00:00 2001
From: Jiayuan Zhang <forrestchang7@gmail.com>
Date: Sun, 15 Feb 2026 19:53:03 +0800
Subject: [PATCH 07/12] fix(agent): pass sessionId to tools for sub-agent
 session tracking
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

toolsOptions spread `options` which had sessionId undefined for
auto-generated sessions. This caused sessions_list and sessions_spawn
to fail with "No session ID available" — breaking sub-agent orchestration.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---
 packages/core/src/agent/runner.ts | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/packages/core/src/agent/runner.ts b/packages/core/src/agent/runner.ts
index c9f81f03..48bc355d 100644
--- a/packages/core/src/agent/runner.ts
+++ b/packages/core/src/agent/runner.ts
@@ -393,9 +393,12 @@ export class Agent {
     const profileToolsConfig = this.profile?.getToolsConfig();
     const mergedToolsConfig = mergeToolsConfig(profileToolsConfig, options.tools);
     const profileDir = this.profile?.getProfileDir();
+    // Use this.sessionId (which may be auto-generated) instead of options.sessionId
+    // (which may be undefined). Without this, sessions_list and sessions_spawn
+    // can't find sub-agent runs because they have no session context.
     this.toolsOptions = mergedToolsConfig
-      ? { ...options, cwd: effectiveCwd, tools: mergedToolsConfig, profileDir, provider: this.resolvedProvider }
-      : { ...options, cwd: effectiveCwd, profileDir, provider: this.resolvedProvider };
+      ? { ...options, sessionId: this.sessionId, cwd: effectiveCwd, tools: mergedToolsConfig, profileDir, provider: this.resolvedProvider }
+      : { ...options, sessionId: this.sessionId, cwd: effectiveCwd, profileDir, provider: this.resolvedProvider };
 
     const tools = resolveTools(this.toolsOptions);
     if (this.debug) {

From 691e33e71ed177c0358eb3db8aa7e3d0731f1604 Mon Sep 17 00:00:00 2001
From: Jiayuan Zhang <forrestchang7@gmail.com>
Date: Sun, 15 Feb 2026 19:53:12 +0800
Subject: [PATCH 08/12] fix(tools): auto-create group when custom groupId is
 provided

LLM often invents custom groupId strings that don't exist in the
registry, causing "group not found" errors. Now auto-creates the
group instead, matching the behavior when `next` is provided.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---
 .../src/agent/tools/sessions-spawn.test.ts    | 23 ++++++++++++++++++-
 .../core/src/agent/tools/sessions-spawn.ts    | 19 +++++++--------
 2 files changed, 32 insertions(+), 10 deletions(-)

diff --git a/packages/core/src/agent/tools/sessions-spawn.test.ts b/packages/core/src/agent/tools/sessions-spawn.test.ts
index 9eaff1a7..0a0d80a9 100644
--- a/packages/core/src/agent/tools/sessions-spawn.test.ts
+++ b/packages/core/src/agent/tools/sessions-spawn.test.ts
@@ -1,7 +1,11 @@
-import { describe, it, expect } from "vitest";
+import { describe, it, expect, beforeEach } from "vitest";
 import { createSessionsSpawnTool } from "./sessions-spawn.js";
+import { getSubagentGroup, resetSubagentRegistryForTests } from "../subagent/registry.js";
 
 describe("sessions_spawn tool", () => {
+  beforeEach(() => {
+    resetSubagentRegistryForTests();
+  });
   it("has correct name and description", () => {
     const tool = createSessionsSpawnTool({ isSubagent: false, sessionId: "test-session" });
     expect(tool.name).toBe("sessions_spawn");
@@ -24,6 +28,23 @@ describe("sessions_spawn tool", () => {
     expect(firstContent.text).toContain("not allowed");
   });
 
+  it("auto-creates group when custom groupId is provided", async () => {
+    const tool = createSessionsSpawnTool({ isSubagent: false, sessionId: "parent-session" });
+
+    // Should not error — the group is auto-created
+    await tool.execute(
+      "call-group",
+      { task: "research topic", label: "Research", groupId: "my-custom-group" } as any,
+      new AbortController().signal,
+    );
+
+    // Verify group was created in the registry
+    const group = getSubagentGroup("my-custom-group");
+    expect(group).toBeDefined();
+    expect(group!.groupId).toBe("my-custom-group");
+    expect(group!.label).toBe("Group: Research");
+  });
+
   it("fails gracefully when Hub is not initialized", async () => {
     const tool = createSessionsSpawnTool({ isSubagent: false, sessionId: "parent-session" });
 
diff --git a/packages/core/src/agent/tools/sessions-spawn.ts b/packages/core/src/agent/tools/sessions-spawn.ts
index 63e615bb..030fbe27 100644
--- a/packages/core/src/agent/tools/sessions-spawn.ts
+++ b/packages/core/src/agent/tools/sessions-spawn.ts
@@ -126,19 +126,20 @@ export function createSessionsSpawnTool(
       const runId = uuidv7();
       const childSessionId = uuidv7();
 
-      // Validate groupId if provided
+      // Auto-create group when groupId is provided but doesn't exist yet,
+      // or when `next` is provided without a groupId.
       if (groupId) {
         const existingGroup = getSubagentGroup(groupId);
         if (!existingGroup) {
-          return {
-            content: [{ type: "text", text: `Error: group not found: ${groupId}. Use the groupId returned by a previous sessions_spawn call.` }],
-            details: { status: "error", error: `group not found: ${groupId}` },
-          };
+          // LLM provided a custom groupId — auto-create the group
+          createSubagentGroup({
+            groupId,
+            requesterSessionId,
+            label: label ? `Group: ${label}` : undefined,
+            next,
+          });
         }
-      }
-
-      // Auto-create group when `next` is provided without an existing groupId
-      if (!groupId && next) {
+      } else if (next) {
         groupId = uuidv7();
         createSubagentGroup({
           groupId,

From b394b0ccf9635261b17b3af88d0d249b63a96bd1 Mon Sep 17 00:00:00 2001
From: Jiayuan Zhang <forrestchang7@gmail.com>
Date: Sun, 15 Feb 2026 19:53:18 +0800
Subject: [PATCH 09/12] fix(skills): use python3 and inject skill directory
 path into prompt
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- SKILL.md: python → python3 (macOS has no `python` binary)
- skills/index.ts: inject skill directory path so agent can resolve
  relative paths like scripts/recalc.py to absolute paths

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---
 packages/core/src/agent/skills/index.ts | 6 ++++++
 skills/xlsx/SKILL.md                    | 6 +++---
 2 files changed, 9 insertions(+), 3 deletions(-)

diff --git a/packages/core/src/agent/skills/index.ts b/packages/core/src/agent/skills/index.ts
index e7343eb2..0a84f076 100644
--- a/packages/core/src/agent/skills/index.ts
+++ b/packages/core/src/agent/skills/index.ts
@@ -5,6 +5,7 @@
  * Compatible with OpenClaw/AgentSkills specification
  */
 
+import { dirname } from "path";
 import type { Skill, SkillManagerOptions, SkillsConfig, SkillCommandSpec, SkillInvocationResult } from "./types.js";
 import { loadAllSkills, getProfileSkillsDir, initializeManagedSkills, getManagedSkillsDir } from "./loader.js";
 import {
@@ -332,6 +333,11 @@ export class SkillManager {
         parts.push(`## ${emoji} ${name} (${id})`);
         parts.push(`${desc}\n`);
 
+        // Include skill directory path so the agent can resolve relative paths
+        // (e.g., scripts/recalc.py → /absolute/path/to/skill/scripts/recalc.py)
+        const skillDir = dirname(skill.filePath);
+        parts.push(`**Skill directory**: \`${skillDir}\`\n`);
+
         // Include full instructions
         if (skill.instructions) {
           parts.push(skill.instructions);
diff --git a/skills/xlsx/SKILL.md b/skills/xlsx/SKILL.md
index 8a617e58..dfdb1f6a 100644
--- a/skills/xlsx/SKILL.md
+++ b/skills/xlsx/SKILL.md
@@ -151,7 +151,7 @@ This applies to ALL calculations - totals, percentages, ratios, differences, etc
 4. **Save**: Write to file
 5. **Recalculate formulas (MANDATORY IF USING FORMULAS)**: Use the scripts/recalc.py script
    ```bash
-   python scripts/recalc.py output.xlsx
+   python3 scripts/recalc.py output.xlsx
    ```
 6. **Verify and fix any errors**:
    - The script returns JSON with error details
@@ -224,12 +224,12 @@ wb.save('modified.xlsx')
 Excel files created or modified by openpyxl contain formulas as strings but not calculated values. Use the provided `scripts/recalc.py` script to recalculate formulas:
 
 ```bash
-python scripts/recalc.py <excel_file> [timeout_seconds]
+python3 scripts/recalc.py <excel_file> [timeout_seconds]
 ```
 
 Example:
 ```bash
-python scripts/recalc.py output.xlsx 30
+python3 scripts/recalc.py output.xlsx 30
 ```
 
 The script:

From e39f9a5dfe510261e13805754fcc8d8ad1e68361 Mon Sep 17 00:00:00 2001
From: Jiayuan Zhang <forrestchang7@gmail.com>
Date: Sun, 15 Feb 2026 19:53:26 +0800
Subject: [PATCH 10/12] feat(cli): wait for sub-agents and output findings in
 run mode

In CLI mode, the parent Agent is not registered with the Hub, so the
normal announce flow can't deliver sub-agent results. Added polling
mechanism that waits for sub-agents to complete and prints their
findings directly to stdout.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---
 apps/cli/src/commands/run.ts | 47 +++++++++++++++++++++++++++++++++++-
 1 file changed, 46 insertions(+), 1 deletion(-)

diff --git a/apps/cli/src/commands/run.ts b/apps/cli/src/commands/run.ts
index c5fb6916..157baeab 100644
--- a/apps/cli/src/commands/run.ts
+++ b/apps/cli/src/commands/run.ts
@@ -7,7 +7,7 @@
  */
 
 import { join } from "node:path";
-import { Agent, Hub } from "@multica/core";
+import { Agent, Hub, listSubagentRuns } from "@multica/core";
 import type { AgentOptions } from "@multica/core";
 import type { ToolsConfig } from "@multica/core";
 import { DATA_DIR } from "@multica/utils";
@@ -230,7 +230,52 @@ export async function runCommand(args: string[]): Promise<void> {
       console.error(`Error: ${result.error}`);
       process.exitCode = 1;
     }
+
+    // Wait for sub-agents to complete and parent to process their results.
+    // Without this, CLI exits before sub-agent announcements are delivered.
+    await waitForSubagents(agent);
   } finally {
     hub.shutdown();
   }
 }
+
+/**
+ * Wait for any running sub-agents to complete, then output their findings.
+ *
+ * In CLI mode, the parent Agent is not registered with the Hub, so the normal
+ * announce flow (Hub → writeInternal) can't deliver results. Instead, we poll
+ * the registry and print findings directly once all sub-agents finish.
+ *
+ * Max wait: 30 minutes (matches default sub-agent timeout).
+ */
+async function waitForSubagents(agent: Agent): Promise<void> {
+  const MAX_WAIT_MS = 30 * 60 * 1000;
+  const POLL_INTERVAL_MS = 2000;
+  const start = Date.now();
+
+  const allRuns = listSubagentRuns(agent.sessionId);
+  if (allRuns.length === 0) return;
+
+  // Phase 1: Wait for all sub-agent runs to finish
+  while (Date.now() - start < MAX_WAIT_MS) {
+    const runs = listSubagentRuns(agent.sessionId);
+    const running = runs.filter((r) => !r.endedAt);
+    if (running.length === 0) break;
+    console.error(dim(`[waiting for ${running.length} sub-agent(s)...]`));
+    await new Promise((r) => setTimeout(r, POLL_INTERVAL_MS));
+  }
+
+  // Phase 2: Output sub-agent findings directly (bypasses Hub announce flow)
+  const completedRuns = listSubagentRuns(agent.sessionId).filter((r) => r.endedAt);
+  if (completedRuns.length === 0) return;
+
+  console.error(dim(`[${completedRuns.length} sub-agent(s) completed]`));
+
+  for (const run of completedRuns) {
+    const displayName = run.label || run.task.slice(0, 60);
+    const status = run.outcome?.status ?? "unknown";
+    const findings = run.findings || "(no output)";
+    console.log(`\n--- Sub-agent: ${displayName} [${status}] ---`);
+    console.log(findings);
+  }
+}

From 084657868fff96d207789d88c1a212833d0b51b6 Mon Sep 17 00:00:00 2001
From: Jiayuan Zhang <forrestchang7@gmail.com>
Date: Sun, 15 Feb 2026 20:43:37 +0800
Subject: [PATCH 11/12] revert(agent): remove parallel tool execution patch,
 keep serial

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---
 package.json                      |  3 ---
 packages/core/src/agent/runner.ts |  8 +++-----
 pnpm-lock.yaml                    | 15 +++++----------
 3 files changed, 8 insertions(+), 18 deletions(-)

diff --git a/package.json b/package.json
index d39e6145..80b868f4 100644
--- a/package.json
+++ b/package.json
@@ -44,9 +44,6 @@
     "overrides": {
       "@types/react": "catalog:",
       "@types/react-dom": "catalog:"
-    },
-    "patchedDependencies": {
-      "@mariozechner/pi-agent-core@0.52.9": "patches/@mariozechner__pi-agent-core@0.52.9.patch"
     }
   },
   "devDependencies": {
diff --git a/packages/core/src/agent/runner.ts b/packages/core/src/agent/runner.ts
index 48bc355d..da3112a1 100644
--- a/packages/core/src/agent/runner.ts
+++ b/packages/core/src/agent/runner.ts
@@ -783,19 +783,17 @@ export class Agent {
 
   private handleRunLogEvent(event: AgentEvent) {
     if (event.type === "tool_execution_start") {
-      const toolCallId = (event as any).toolCallId ?? "unknown";
       const toolName = (event as any).toolName ?? "unknown";
-      this.toolStartTimes.set(toolCallId, Date.now());
+      this.toolStartTimes.set(toolName, Date.now());
       this.runLog.log("tool_start", {
         tool: toolName,
         args: JSON.stringify((event as any).args ?? {}).slice(0, 500),
       });
     } else if (event.type === "tool_execution_end") {
-      const toolCallId = (event as any).toolCallId ?? "unknown";
       const toolName = (event as any).toolName ?? "unknown";
-      const startTime = this.toolStartTimes.get(toolCallId);
+      const startTime = this.toolStartTimes.get(toolName);
       const duration_ms = startTime ? Date.now() - startTime : undefined;
-      this.toolStartTimes.delete(toolCallId);
+      this.toolStartTimes.delete(toolName);
 
       // Extract result metadata for run-log persistence (survives session compaction)
       const result = (event as any).result;
diff --git a/pnpm-lock.yaml b/pnpm-lock.yaml
index 18c13e8b..cc765ee9 100644
--- a/pnpm-lock.yaml
+++ b/pnpm-lock.yaml
@@ -68,18 +68,13 @@ overrides:
   '@types/react': ^19.2.0
   '@types/react-dom': ^19.2.0
 
-patchedDependencies:
-  '@mariozechner/pi-agent-core@0.52.9':
-    hash: befbe3b9fd1a6d3e13dfa4927d9f0addd2f3b3454bd9d407c16dacb9a71c0f00
-    path: patches/@mariozechner__pi-agent-core@0.52.9.patch
-
 importers:
 
   .:
     dependencies:
       '@mariozechner/pi-agent-core':
         specifier: 'catalog:'
-        version: 0.52.9(patch_hash=befbe3b9fd1a6d3e13dfa4927d9f0addd2f3b3454bd9d407c16dacb9a71c0f00)(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
+        version: 0.52.9(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
       '@mariozechner/pi-ai':
         specifier: 'catalog:'
         version: 0.52.9(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
@@ -596,7 +591,7 @@ importers:
     dependencies:
       '@mariozechner/pi-agent-core':
         specifier: 'catalog:'
-        version: 0.52.9(patch_hash=befbe3b9fd1a6d3e13dfa4927d9f0addd2f3b3454bd9d407c16dacb9a71c0f00)(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
+        version: 0.52.9(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
       '@mariozechner/pi-ai':
         specifier: 'catalog:'
         version: 0.52.9(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
@@ -695,7 +690,7 @@ importers:
     devDependencies:
       '@mariozechner/pi-agent-core':
         specifier: 'catalog:'
-        version: 0.52.9(patch_hash=befbe3b9fd1a6d3e13dfa4927d9f0addd2f3b3454bd9d407c16dacb9a71c0f00)(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
+        version: 0.52.9(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
       '@mariozechner/pi-ai':
         specifier: 'catalog:'
         version: 0.52.9(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
@@ -12503,7 +12498,7 @@ snapshots:
       std-env: 3.10.0
       yoctocolors: 2.1.2
 
-  '@mariozechner/pi-agent-core@0.52.9(patch_hash=befbe3b9fd1a6d3e13dfa4927d9f0addd2f3b3454bd9d407c16dacb9a71c0f00)(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)':
+  '@mariozechner/pi-agent-core@0.52.9(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)':
     dependencies:
       '@mariozechner/pi-ai': 0.52.9(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
     transitivePeerDependencies:
@@ -12542,7 +12537,7 @@ snapshots:
   '@mariozechner/pi-coding-agent@0.52.9(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)':
     dependencies:
       '@mariozechner/jiti': 2.6.5
-      '@mariozechner/pi-agent-core': 0.52.9(patch_hash=befbe3b9fd1a6d3e13dfa4927d9f0addd2f3b3454bd9d407c16dacb9a71c0f00)(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
+      '@mariozechner/pi-agent-core': 0.52.9(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
       '@mariozechner/pi-ai': 0.52.9(@modelcontextprotocol/sdk@1.26.0(zod@4.3.6))(ws@8.19.0)(zod@4.3.6)
       '@mariozechner/pi-tui': 0.52.9
       '@silvia-odwyer/photon-node': 0.3.4

From a443d3009d5b1c761e0b03247d5a6e9c02bfb956 Mon Sep 17 00:00:00 2001
From: Jiayuan Zhang <forrestchang7@gmail.com>
Date: Sun, 15 Feb 2026 20:45:32 +0800
Subject: [PATCH 12/12] chore: remove leftover patch file from parallel
 execution revert

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---
 .../@mariozechner__pi-agent-core@0.52.9.patch | 80 -------------------
 1 file changed, 80 deletions(-)
 delete mode 100644 patches/@mariozechner__pi-agent-core@0.52.9.patch

diff --git a/patches/@mariozechner__pi-agent-core@0.52.9.patch b/patches/@mariozechner__pi-agent-core@0.52.9.patch
deleted file mode 100644
index e570a653..00000000
--- a/patches/@mariozechner__pi-agent-core@0.52.9.patch
+++ /dev/null
@@ -1,80 +0,0 @@
-diff --git a/dist/agent-loop.js b/dist/agent-loop.js
-index b26d45753809da14df6409354e8c537684b9acd7..931399cc12bafea940fe99ab74ae288d71506e52 100644
---- a/dist/agent-loop.js
-+++ b/dist/agent-loop.js
-@@ -206,17 +206,18 @@ async function streamAssistantResponse(context, config, signal, stream, streamFn
-  */
- async function executeToolCalls(tools, assistantMessage, signal, stream, getSteeringMessages) {
-     const toolCalls = assistantMessage.content.filter((c) => c.type === "toolCall");
--    const results = [];
--    let steeringMessages;
--    for (let index = 0; index < toolCalls.length; index++) {
--        const toolCall = toolCalls[index];
--        const tool = tools?.find((t) => t.name === toolCall.name);
-+    // Emit all tool_execution_start events immediately
-+    for (const toolCall of toolCalls) {
-         stream.push({
-             type: "tool_execution_start",
-             toolCallId: toolCall.id,
-             toolName: toolCall.name,
-             args: toolCall.arguments,
-         });
-+    }
-+    // Execute all tools in parallel
-+    const settled = await Promise.allSettled(toolCalls.map(async (toolCall) => {
-+        const tool = tools?.find((t) => t.name === toolCall.name);
-         let result;
-         let isError = false;
-         try {
-@@ -240,6 +241,27 @@ async function executeToolCalls(tools, assistantMessage, signal, stream, getStee
-             };
-             isError = true;
-         }
-+        return { result, isError };
-+    }));
-+    // Process results IN ORIGINAL ORDER (critical for LLM context)
-+    const results = [];
-+    let steeringMessages;
-+    for (let i = 0; i < settled.length; i++) {
-+        const entry = settled[i];
-+        const toolCall = toolCalls[i];
-+        let result;
-+        let isError;
-+        if (entry.status === "fulfilled") {
-+            result = entry.value.result;
-+            isError = entry.value.isError;
-+        }
-+        else {
-+            result = {
-+                content: [{ type: "text", text: entry.reason instanceof Error ? entry.reason.message : String(entry.reason) }],
-+                details: {},
-+            };
-+            isError = true;
-+        }
-         stream.push({
-             type: "tool_execution_end",
-             toolCallId: toolCall.id,
-@@ -259,17 +281,12 @@ async function executeToolCalls(tools, assistantMessage, signal, stream, getStee
-         results.push(toolResultMessage);
-         stream.push({ type: "message_start", message: toolResultMessage });
-         stream.push({ type: "message_end", message: toolResultMessage });
--        // Check for steering messages - skip remaining tools if user interrupted
--        if (getSteeringMessages) {
--            const steering = await getSteeringMessages();
--            if (steering.length > 0) {
--                steeringMessages = steering;
--                const remainingCalls = toolCalls.slice(index + 1);
--                for (const skipped of remainingCalls) {
--                    results.push(skipToolCall(skipped, stream));
--                }
--                break;
--            }
-+    }
-+    // Check steering messages once after all tools complete
-+    if (getSteeringMessages) {
-+        const steering = await getSteeringMessages();
-+        if (steering.length > 0) {
-+            steeringMessages = steering;
-         }
-     }
-     return { toolResults: results, steeringMessages };