muonroi-cli 1.7.2 → 1.8.0

package/dist/src/models/catalog-gemini.test.js

@@ -2,17 +2,17 @@ import { beforeAll, describe, expect, it } from "vitest";
 import { detectProviderForModel } from "../providers/runtime.js";
 import { getModelInfo, loadCatalog } from "./registry.js";
 /**
- * Regression guard for the Gemini catalog entries (added 2026-06-04). Model IDs
- * verified against ai.google.dev/gemini-api/docs/models on that date. Without a
- * catalog entry a model cannot be selected or routed (Zero-Hardcode Rule), so
- * these assertions lock in that Gemini is actually usable end-to-end.
+ * Regression guard for the Agy/Google Gemini catalog entries.
+ * These are the models Agy supports for the "google" provider (via agy.exe OAuth + cloudcode-pa).
+ * Without a catalog entry a model cannot be selected or routed (Zero-Hardcode Rule).
  */
 const GEMINI_IDS = [
-    "gemini-2.5-flash",
-    "gemini-2.5-pro",
-    "gemini-3.5-flash",
-    "gemini-3.1-flash-lite",
-    "gemini-3.1-pro-preview",
+    "gemini-3.5-flash-high",
+    "gemini-3.5-flash-medium",
+    "gemini-3.5-flash-low",
+    "gemini-3.1-pro-high",
+    "gemini-3.1-pro-low",
+    "gemini-3-flash",
 ];
 describe("Gemini catalog entries", () => {
     beforeAll(async () => {
@@ -26,8 +26,10 @@ describe("Gemini catalog entries", () => {
         }
     });
     it("maps gemini ids to the google provider via runtime detection", () => {
-        expect(detectProviderForModel("gemini-2.5-flash")).toBe("google");
+        // direct + alias (gemini-3.5-flash aliases to the high variant)
+        expect(detectProviderForModel("gemini-3.5-flash-high")).toBe("google");
         expect(detectProviderForModel("gemini-3.5-flash")).toBe("google");
+        expect(detectProviderForModel("gemini-3.1-pro-high")).toBe("google");
     });
     it("marks all Gemini models multimodal (vision) and 1M context", () => {
         for (const id of GEMINI_IDS) {

package/dist/src/models/catalog.json

@@ -1,7 +1,7 @@
 {
   "version": "2.2",
   "updated_at": "2026-06-04",
-  "description": "Local fallback catalog: deepseek + siliconflow + openai (ChatGPT OAuth) + google (Gemini, via AI Studio key or Google OAuth). SiliconFlow pricing verified 2026-05-21; Gemini 2.5 pricing per ai.google.dev/pricing 2026-06-04 (3.x flash/pro entries APPROXIMATE — see per-model descriptions). Model IDs verified against ai.google.dev/gemini-api/docs/models 2026-06-04. See ./catalog.README.md for curation rationale.",
+  "description": "Local fallback catalog: deepseek + siliconflow + openai (ChatGPT OAuth) + google (Agy via agy.exe OAuth + cloudcode-pa). Agy supports: Gemini 3.5 Flash (High/Medium/Low), Gemini 3.1 Pro (High/Low), Gemini 3 Flash, Claude Sonnet 4.6 (Thinking), Claude Opus 4.6 (Thinking), GPT-OSS 120B (Medium+). Non-Gemini models under google provider use Agy token for proxied access. Other providers unchanged. Pricing APPROXIMATE.",
   "models": [
     {
       "id": "deepseek-v4-flash",
@@ -227,42 +227,42 @@
       "supports_vision": false
     },
     {
-      "id": "gemini-2.5-flash",
-      "name": "Gemini 2.5 Flash",
+      "id": "gemini-3.5-flash-high",
+      "name": "Gemini 3.5 Flash (High)",
       "provider": "google",
-      "tier": "balanced",
+      "tier": "fast",
       "context_window": 1048576,
       "max_output_tokens": 65536,
-      "input_price_per_million": 0.3,
-      "output_price_per_million": 2.5,
+      "input_price_per_million": 0.5,
+      "output_price_per_million": 3.0,
       "reasoning": true,
       "thinking_type": "enabled",
-      "supports_effort": false,
-      "default_reasoning_effort": null,
-      "description": "Google Gemini 2.5 Flash via Generative Language API (AI Studio key — console aistudio.google.com/app/apikey, or stored Google OAuth). Multimodal, 1M context, thinking-capable. Pricing per ai.google.dev/pricing (2026-06-04).",
-      "aliases": ["gemini-flash", "gemini-2.5-flash-latest"],
+      "supports_effort": true,
+      "default_reasoning_effort": "high",
+      "description": "Agy-supported Gemini 3.5 Flash High via Agy OAuth (cloudcode-pa backend, not public Generative Language). Multimodal, 1M context. Matches official agy cli model list.",
+      "aliases": ["gemini-3.5-flash"],
       "supports_vision": true
     },
     {
-      "id": "gemini-2.5-pro",
-      "name": "Gemini 2.5 Pro",
+      "id": "gemini-3.5-flash-medium",
+      "name": "Gemini 3.5 Flash (Medium)",
       "provider": "google",
-      "tier": "premium",
+      "tier": "balanced",
       "context_window": 1048576,
       "max_output_tokens": 65536,
-      "input_price_per_million": 1.25,
-      "output_price_per_million": 10.0,
+      "input_price_per_million": 0.5,
+      "output_price_per_million": 3.0,
       "reasoning": true,
       "thinking_type": "enabled",
-      "supports_effort": false,
-      "default_reasoning_effort": null,
-      "description": "Google Gemini 2.5 Pro via Generative Language API (AI Studio key or stored Google OAuth). Most advanced 2.5 reasoning model, multimodal, 1M context. Pricing per ai.google.dev/pricing (2026-06-04).",
-      "aliases": ["gemini-pro", "gemini-2.5-pro-latest"],
+      "supports_effort": true,
+      "default_reasoning_effort": "medium",
+      "description": "Agy-supported Gemini 3.5 Flash Medium via Agy OAuth (cloudcode-pa). Multimodal, 1M context. Matches official agy cli model list.",
+      "aliases": ["gemini-3.5-flash-med"],
       "supports_vision": true
     },
     {
-      "id": "gemini-3.5-flash",
-      "name": "Gemini 3.5 Flash",
+      "id": "gemini-3.5-flash-low",
+      "name": "Gemini 3.5 Flash (Low)",
       "provider": "google",
       "tier": "fast",
       "context_window": 1048576,
@@ -271,32 +271,32 @@
       "output_price_per_million": 3.0,
       "reasoning": true,
       "thinking_type": "enabled",
-      "supports_effort": false,
-      "default_reasoning_effort": null,
-      "description": "Google Gemini 3.5 Flash via Generative Language API. Frontier agentic/coding flash model, multimodal, 1M context, thinking-capable. Pricing APPROXIMATE (verify at ai.google.dev/pricing before cost-sensitive use).",
-      "aliases": ["gemini-flash-3.5"],
+      "supports_effort": true,
+      "default_reasoning_effort": "low",
+      "description": "Agy-supported Gemini 3.5 Flash Low via Agy OAuth (cloudcode-pa). Multimodal, 1M context. Matches official agy cli model list.",
+      "aliases": ["gemini-3.5-flash-low"],
       "supports_vision": true
     },
     {
-      "id": "gemini-3.1-flash-lite",
-      "name": "Gemini 3.1 Flash-Lite",
+      "id": "gemini-3.1-pro-high",
+      "name": "Gemini 3.1 Pro (High)",
       "provider": "google",
-      "tier": "fast",
+      "tier": "premium",
       "context_window": 1048576,
       "max_output_tokens": 65536,
-      "input_price_per_million": 0.15,
-      "output_price_per_million": 0.6,
+      "input_price_per_million": 2.0,
+      "output_price_per_million": 12.0,
       "reasoning": true,
       "thinking_type": "enabled",
-      "supports_effort": false,
-      "default_reasoning_effort": null,
-      "description": "Google Gemini 3.1 Flash-Lite via Generative Language API. Cheapest multimodal Gemini, 1M context, limited thinking. Pricing APPROXIMATE (verify at ai.google.dev/pricing before cost-sensitive use).",
-      "aliases": ["gemini-flash-lite"],
+      "supports_effort": true,
+      "default_reasoning_effort": "high",
+      "description": "Agy-supported Gemini 3.1 Pro High via Agy OAuth (cloudcode-pa backend). Strongest agentic Gemini. Matches official agy cli model list. Pricing approximate.",
+      "aliases": ["gemini-3.1-pro", "gemini-3.1-pro-preview"],
       "supports_vision": true
     },
     {
-      "id": "gemini-3.1-pro-preview",
-      "name": "Gemini 3.1 Pro Preview",
+      "id": "gemini-3.1-pro-low",
+      "name": "Gemini 3.1 Pro (Low)",
       "provider": "google",
       "tier": "premium",
       "context_window": 1048576,
@@ -305,12 +305,80 @@
       "output_price_per_million": 12.0,
       "reasoning": true,
       "thinking_type": "enabled",
+      "supports_effort": true,
+      "default_reasoning_effort": "low",
+      "description": "Agy-supported Gemini 3.1 Pro Low via Agy OAuth (cloudcode-pa). Matches official agy cli model list.",
+      "aliases": ["gemini-3.1-pro-low"],
+      "supports_vision": true
+    },
+    {
+      "id": "gemini-3-flash",
+      "name": "Gemini 3 Flash",
+      "provider": "google",
+      "tier": "fast",
+      "context_window": 1048576,
+      "max_output_tokens": 65536,
+      "input_price_per_million": 0.3,
+      "output_price_per_million": 2.0,
+      "reasoning": true,
+      "thinking_type": "enabled",
       "supports_effort": false,
       "default_reasoning_effort": null,
-      "description": "Google Gemini 3.1 Pro Preview via Generative Language API. Strongest agentic/reasoning Gemini, multimodal, 1M context. Preview model — id/pricing may change; pricing APPROXIMATE (verify at ai.google.dev/pricing).",
-      "aliases": ["gemini-pro-preview"],
+      "description": "Agy-supported Gemini 3 Flash via Agy OAuth (cloudcode-pa). Multimodal. Matches official agy cli model list.",
+      "aliases": ["gemini-3-flash"],
+      "supports_vision": true
+    },
+    {
+      "id": "claude-sonnet-4.6-thinking",
+      "name": "Claude Sonnet 4.6 (Thinking)",
+      "provider": "google",
+      "tier": "premium",
+      "context_window": 200000,
+      "max_output_tokens": 32000,
+      "input_price_per_million": 3.0,
+      "output_price_per_million": 15.0,
+      "reasoning": true,
+      "thinking_type": "enabled",
+      "supports_effort": true,
+      "default_reasoning_effort": "medium",
+      "description": "Agy-supported Claude Sonnet 4.6 Thinking via Agy OAuth (cloudcode-pa backend proxy). Matches official agy cli model list.",
+      "aliases": ["claude-sonnet-4-6-thinking", "claude-4.6-sonnet-thinking"],
+      "supports_vision": true
+    },
+    {
+      "id": "claude-opus-4.6-thinking",
+      "name": "Claude Opus 4.6 (Thinking)",
+      "provider": "google",
+      "tier": "premium",
+      "context_window": 200000,
+      "max_output_tokens": 32000,
+      "input_price_per_million": 15.0,
+      "output_price_per_million": 75.0,
+      "reasoning": true,
+      "thinking_type": "enabled",
+      "supports_effort": true,
+      "default_reasoning_effort": "medium",
+      "description": "Agy-supported Claude Opus 4.6 Thinking via Agy OAuth (cloudcode-pa backend proxy). Matches official agy cli model list.",
+      "aliases": ["claude-opus-4-6-thinking", "claude-4.6-opus-thinking"],
       "supports_vision": true
     },
+    {
+      "id": "gpt-oss-120b-medium",
+      "name": "GPT-OSS 120B (Medium)",
+      "provider": "google",
+      "tier": "balanced",
+      "context_window": 128000,
+      "max_output_tokens": 32000,
+      "input_price_per_million": 0.2,
+      "output_price_per_million": 0.8,
+      "reasoning": true,
+      "thinking_type": "enabled",
+      "supports_effort": true,
+      "default_reasoning_effort": "medium",
+      "description": "Agy-supported GPT-OSS 120B Medium via Agy OAuth (cloudcode-pa). Matches official agy cli model list.",
+      "aliases": ["gpt-oss-120b", "gpt-oss-medium"],
+      "supports_vision": false
+    },
     {
       "id": "grok-4.3",
       "name": "Grok 4.3 (xAI)",

package/dist/src/ops/doctor.js

@@ -169,9 +169,9 @@ async function checkBrainEmptiness() {
         // Count ee_injection events with event_subtype='no_match' in last 30 days
         const cutoff = new Date(Date.now() - 30 * 86_400_000).toISOString();
         const row = db
-            .prepare(`SELECT COUNT(*) as cnt FROM interaction_logs
-         WHERE event_type = 'ee_injection'
-           AND event_subtype = 'no_match'
+            .prepare(`SELECT COUNT(*) as cnt FROM interaction_logs
+         WHERE event_type = 'ee_injection'
+           AND event_subtype = 'no_match'
            AND created_at >= ?`)
             .get(cutoff);
         const noMatchCount = row?.cnt ?? 0;
@@ -293,10 +293,10 @@ async function checkCouncilMcpNudge() {
         // 2. Query DB for [Council Memory] records with URL or research topics
         const db = getDatabase();
         const rows = db
-            .prepare(`SELECT message_json FROM messages
-         WHERE role = 'system'
-           AND message_json LIKE '%[Council Memory]%'
-         ORDER BY created_at DESC
+            .prepare(`SELECT message_json FROM messages
+         WHERE role = 'system'
+           AND message_json LIKE '%[Council Memory]%'
+         ORDER BY created_at DESC
          LIMIT 50`)
             .all();
         let qualifyingCount = 0;

package/dist/src/orchestrator/__tests__/auto-commit.test.d.ts

@@ -0,0 +1 @@
+export {};

package/dist/src/orchestrator/__tests__/auto-commit.test.js

@@ -0,0 +1,142 @@
+import { execFileSync } from "node:child_process";
+import { mkdtempSync, rmSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join, resolve } from "node:path";
+import { afterEach, beforeEach, describe, expect, it } from "vitest";
+import { blockingErrorsForFile, buildFileListSubject, isAutoCommitEnabled, isCliArtifactPath, isCommitGateEnabled, isExcludedPath, isSensitivePath, parsePorcelainPaths, pathsForCommitGate, splitCommitMessage, } from "../auto-commit.js";
+function diag(severity) {
+    return {
+        message: "x",
+        severity,
+        range: { start: { line: 0, character: 0 }, end: { line: 0, character: 1 } },
+    };
+}
+function diagFile(filePath, severities) {
+    return { filePath, serverId: "ts", diagnostics: severities.map(diag) };
+}
+describe("auto-commit pure helpers", () => {
+    it("parses porcelain output incl. rename + quoted paths", () => {
+        const out = ' M src/a.ts\n?? new file.txt\nR  old.ts -> src/b.ts\nA  "with space.ts"\n';
+        const set = parsePorcelainPaths(out);
+        expect(set.has("src/a.ts")).toBe(true);
+        expect(set.has("new file.txt")).toBe(true);
+        // rename keeps ONLY the new path
+        expect(set.has("src/b.ts")).toBe(true);
+        expect(set.has("old.ts")).toBe(false);
+        // quoted path with a space is unquoted
+        expect(set.has("with space.ts")).toBe(true);
+    });
+    it("flags sensitive paths and lets normal source through", () => {
+        expect(isSensitivePath(".env")).toBe(true);
+        expect(isSensitivePath(".env.local")).toBe(true);
+        expect(isSensitivePath("config/app.key")).toBe(true);
+        expect(isSensitivePath("certs/server.pem")).toBe(true);
+        expect(isSensitivePath("src/db-secret.ts")).toBe(true);
+        // .muonroi-cli/ is categorized as a CLI artifact (excluded), not a "secret".
+        expect(isSensitivePath(".muonroi-cli/state.json")).toBe(false);
+        expect(isExcludedPath(".muonroi-cli/state.json")).toBe(true);
+        expect(isSensitivePath("src/index.ts")).toBe(false);
+        expect(isSensitivePath("README.md")).toBe(false);
+    });
+    it("excludes CLI-generated artifacts + build junk (the 11-file-leak fix)", () => {
+        expect(isCliArtifactPath(".muonroi-flow/state.md")).toBe(true);
+        expect(isCliArtifactPath(".muonroi-flow/runs/abc/manifest.md")).toBe(true);
+        expect(isCliArtifactPath(".muonroi-cli/session.db")).toBe(true);
+        expect(isCliArtifactPath(".muonroi-harness-roots.json")).toBe(true);
+        expect(isCliArtifactPath("node_modules/x/index.js")).toBe(true);
+        expect(isCliArtifactPath("dist/bundle.js")).toBe(true);
+        expect(isCliArtifactPath("debug.log")).toBe(true);
+        expect(isCliArtifactPath("src/greeting.txt")).toBe(false);
+        // isExcludedPath unions secrets + artifacts
+        expect(isExcludedPath(".env")).toBe(true);
+        expect(isExcludedPath(".muonroi-flow/state.md")).toBe(true);
+        expect(isExcludedPath("greeting.txt")).toBe(false);
+    });
+    it("backstop subject lists files + stays bounded (never the raw prompt)", () => {
+        expect(buildFileListSubject(["src/a.ts", "src/b.ts"])).toBe("chore: update 2 file(s) — a.ts, b.ts");
+        expect(buildFileListSubject(["a.ts", "b.ts", "c.ts", "d.ts", "e.ts"])).toContain("+2 more");
+        expect(buildFileListSubject(Array.from({ length: 20 }, (_, i) => `file${i}.ts`)).length).toBeLessThanOrEqual(72);
+    });
+    it("splits an agent-authored message: subject <=72, body kept, attribution deduped", () => {
+        expect(splitCommitMessage("feat: add X\r\n\r\nbody line")).toEqual({ subject: "feat: add X", body: "body line" });
+        expect(splitCommitMessage("x".repeat(200)).subject.length).toBeLessThanOrEqual(72);
+        // an attribution the agent already added is dropped (we append exactly one)
+        expect(splitCommitMessage("feat: y\nCoding by - Muonroi-CLI")).toEqual({ subject: "feat: y", body: "" });
+    });
+    it("is disabled under the unit-test runner so the suite never commits", () => {
+        // VITEST is set while this runs → must be false regardless of MUONROI_AUTO_COMMIT.
+        expect(isAutoCommitEnabled()).toBe(false);
+    });
+});
+describe("G1 commit quality gate", () => {
+    it("commit gate is disabled under the unit-test runner (specs commit fixtures freely)", () => {
+        // Mirrors isAutoCommitEnabled — VITEST is set, so the gate must be off here.
+        expect(isCommitGateEnabled()).toBe(false);
+    });
+    describe("blockingErrorsForFile — errors-only + per-file scope", () => {
+        it("returns severity-1 (error) diagnostics for the staged file", () => {
+            const abs = resolve("foo.ts");
+            const errs = blockingErrorsForFile([diagFile("foo.ts", [1, 1])], abs);
+            expect(errs).toHaveLength(2);
+        });
+        it("treats a missing severity as an error (LSP default is 1)", () => {
+            const abs = resolve("foo.ts");
+            expect(blockingErrorsForFile([diagFile("foo.ts", [undefined])], abs)).toHaveLength(1);
+        });
+        it("ignores warnings/infos (severity >= 2) — they never block a commit", () => {
+            const abs = resolve("foo.ts");
+            expect(blockingErrorsForFile([diagFile("foo.ts", [2, 3, 4])], abs)).toHaveLength(0);
+        });
+        it("ignores errors reported against a DIFFERENT file (cross-file breakage doesn't block)", () => {
+            const abs = resolve("foo.ts");
+            // An error on other.ts must not block a commit of foo.ts.
+            expect(blockingErrorsForFile([diagFile("other.ts", [1, 1])], abs)).toHaveLength(0);
+        });
+        it("mixes: keeps only this file's errors out of a multi-file, multi-severity set", () => {
+            const abs = resolve("foo.ts");
+            const errs = blockingErrorsForFile([diagFile("foo.ts", [1, 2, 3]), diagFile("bar.ts", [1, 1])], abs);
+            expect(errs).toHaveLength(1); // only foo.ts's single severity-1
+        });
+        it("empty diagnostics → no blockers (non-source / no-LSP-server files pass)", () => {
+            expect(blockingErrorsForFile([], resolve("README.md"))).toEqual([]);
+        });
+    });
+});
+describe("pathsForCommitGate — bash `git commit` gate path set (real git)", () => {
+    let dir;
+    const g = (args) => execFileSync("git", args, { cwd: dir, stdio: "pipe" });
+    beforeEach(() => {
+        dir = mkdtempSync(join(tmpdir(), "commit-gate-"));
+        g(["init", "-q"]);
+        g(["config", "user.email", "t@t.t"]);
+        g(["config", "user.name", "t"]);
+    });
+    afterEach(() => rmSync(dir, { recursive: true, force: true }));
+    it("returns only the already-staged set for a plain `git commit`", async () => {
+        writeFileSync(join(dir, "a.ts"), "export const a = 1;\n");
+        writeFileSync(join(dir, "b.ts"), "export const b = 2;\n");
+        g(["add", "a.ts"]); // b.ts left untracked
+        const paths = await pathsForCommitGate(dir, { broadAdd: false, commitAll: false });
+        expect(paths).toEqual(["a.ts"]);
+    });
+    it("adds tracked modifications that `git commit -a` will auto-stage", async () => {
+        writeFileSync(join(dir, "a.ts"), "export const a = 1;\n");
+        g(["add", "a.ts"]);
+        g(["commit", "-qm", "init", "--no-verify"]); // temp fixture repo — hook-free by design
+        writeFileSync(join(dir, "a.ts"), "export const a = 99;\n"); // modify tracked, NOT re-staged
+        // plain commit sees nothing staged; commit -a includes the modification.
+        expect(await pathsForCommitGate(dir, { broadAdd: false, commitAll: false })).toEqual([]);
+        expect(await pathsForCommitGate(dir, { broadAdd: false, commitAll: true })).toContain("a.ts");
+    });
+    it("includes the whole working tree for a chained broad `git add -A` (add not yet run)", async () => {
+        writeFileSync(join(dir, "a.ts"), "export const a = 1;\n"); // untracked
+        writeFileSync(join(dir, "c.ts"), "export const c = 3;\n"); // untracked
+        // Nothing staged at pre-exec time → plain set is empty…
+        expect(await pathsForCommitGate(dir, { broadAdd: false, commitAll: false })).toEqual([]);
+        // …but `git add -A` would stage both, so the broad path enumerates them.
+        const broad = await pathsForCommitGate(dir, { broadAdd: true, commitAll: false });
+        expect(broad).toContain("a.ts");
+        expect(broad).toContain("c.ts");
+    });
+});
+//# sourceMappingURL=auto-commit.test.js.map

package/dist/src/orchestrator/__tests__/batch-turn-runner.test.js

@@ -56,6 +56,7 @@ function makeDeps(overrides = {}) {
         runDelegation: async () => ({ success: true, output: "" }),
         readDelegation: async () => ({ success: true, output: "" }),
         listDelegations: async () => ({ success: true, output: "" }),
+        killDelegation: async () => ({ success: true, output: "" }),
         executeBatchToolCall: async (_tools, _call) => ({
             input: {},
             result: { success: true, output: "" },
@@ -78,6 +79,7 @@ function makeArgs(signal, providerStub) {
     const testProviders = getTestProviders();
     return {
         userModelMessage: { role: "user", content: "hello" },
+        userEnrichedMessage: { role: "user", content: "hello" },
         observer: undefined,
         provider: (providerStub ?? {}),
         subagents: [],

package/dist/src/orchestrator/__tests__/delegations.test.d.ts

@@ -0,0 +1 @@
+export {};

package/dist/src/orchestrator/__tests__/delegations.test.js

@@ -0,0 +1,96 @@
+import { createHash } from "crypto";
+import { promises as fs } from "fs";
+import os from "os";
+import path from "path";
+import { describe, expect, it } from "vitest";
+import { DelegationManager } from "../delegations.js";
+function getProjectId(cwd) {
+    const base = path
+        .basename(cwd)
+        .replace(/[^a-zA-Z0-9._-]+/g, "-")
+        .replace(/^-+|-+$/g, "") || "project";
+    const hash = createHash("sha1").update(cwd).digest("hex").slice(0, 10);
+    return `${base}-${hash}`;
+}
+async function getDelegationsDir(cwd) {
+    const projectId = getProjectId(cwd);
+    return path.join(os.homedir(), ".muonroi-cli", "delegations", projectId);
+}
+describe("DelegationManager.kill", () => {
+    it("terminates a running delegation and updates its status to error", async () => {
+        const tempCwd = path.join(os.tmpdir(), "muonroi-test-" + Math.random().toString(36).slice(2));
+        const manager = new DelegationManager(() => tempCwd);
+        const dir = await getDelegationsDir(tempCwd);
+        await fs.mkdir(dir, { recursive: true });
+        const id = "test-amber-panda";
+        const jobPath = path.join(dir, `${id}.json`);
+        const outputPath = path.join(dir, `${id}.md`);
+        // Create a mock running delegation
+        const record = {
+            id,
+            agent: "explore",
+            description: "test delegation",
+            prompt: "noop",
+            cwd: tempCwd,
+            model: "grok-4",
+            sandboxMode: "off",
+            maxToolRounds: 10,
+            maxTokens: 4000,
+            status: "running",
+            startedAt: new Date().toISOString(),
+            outputPath,
+            pid: 99999, // dummy PID
+        };
+        await fs.writeFile(jobPath, JSON.stringify(record, null, 2), "utf8");
+        // Call cancel
+        const result = await manager.kill(id);
+        expect(result.success).toBe(true);
+        expect(result.output).toContain(`Delegation "${id}" (PID: 99999) has been terminated.`);
+        // Verify the record was updated
+        const updatedRaw = await fs.readFile(jobPath, "utf8");
+        const updated = JSON.parse(updatedRaw);
+        expect(updated.status).toBe("error");
+        expect(updated.error).toBe("Cancelled by user.");
+        // Cleanup
+        await fs.rm(jobPath, { force: true });
+        await fs.rm(outputPath, { force: true });
+        await fs.rm(dir, { recursive: true, force: true });
+    });
+    it("returns failure if the delegation ID is not found", async () => {
+        const tempCwd = path.join(os.tmpdir(), "muonroi-test-" + Math.random().toString(36).slice(2));
+        const manager = new DelegationManager(() => tempCwd);
+        const result = await manager.kill("non-existent-id");
+        expect(result.success).toBe(false);
+        expect(result.output).toContain('Delegation "non-existent-id" not found.');
+    });
+    it("returns failure if the delegation is already finished", async () => {
+        const tempCwd = path.join(os.tmpdir(), "muonroi-test-" + Math.random().toString(36).slice(2));
+        const manager = new DelegationManager(() => tempCwd);
+        const dir = await getDelegationsDir(tempCwd);
+        await fs.mkdir(dir, { recursive: true });
+        const id = "test-steady-owl";
+        const jobPath = path.join(dir, `${id}.json`);
+        const record = {
+            id,
+            agent: "explore",
+            description: "test delegation",
+            prompt: "noop",
+            cwd: tempCwd,
+            model: "grok-4",
+            sandboxMode: "off",
+            maxToolRounds: 10,
+            maxTokens: 4000,
+            status: "complete",
+            startedAt: new Date().toISOString(),
+            outputPath: path.join(dir, `${id}.md`),
+        };
+        await fs.writeFile(jobPath, JSON.stringify(record, null, 2), "utf8");
+        const result = await manager.kill(id);
+        expect(result.success).toBe(false);
+        expect(result.output).toContain(`Delegation "${id}" is not running (status: complete).`);
+        // Cleanup
+        await fs.rm(jobPath, { force: true });
+        await fs.rm(dir, { recursive: true, force: true });
+    });
+});
+//# sourceMappingURL=delegations.test.js.map

package/dist/src/orchestrator/__tests__/message-processor.test.js

@@ -99,6 +99,7 @@ function makeDeps(overrides = {}) {
         runDelegation: async () => ({ success: true, output: "" }),
         readDelegation: async () => ({ success: true, output: "" }),
         listDelegations: async () => ({ success: true, output: "" }),
+        killDelegation: async () => ({ success: true, output: "" }),
         appendCompletedTurn: () => { },
         discardAbortedTurn: () => { },
         recordUsage: () => { },

package/dist/src/orchestrator/__tests__/stream-runner.test.js

@@ -169,5 +169,34 @@ describe("StreamRunner — DI surface", () => {
         const b = new StreamRunner(makeDeps());
         expect(a).not.toBe(b);
     });
+    describe("StreamRunner — maxSteps custom overrides", () => {
+        it("respects default maxSteps logic when request.maxToolRounds is undefined", async () => {
+            const deps = makeDeps({ getMaxToolRounds: () => 50 });
+            const runner = new StreamRunner(deps);
+            const outcome = await runner.setup({
+                agent: "explore",
+                description: "test",
+                prompt: "test",
+            });
+            expect(outcome.kind).toBe("prepared");
+            if (outcome.kind === "prepared") {
+                expect(outcome.prepared.maxSteps).toBe(100); // deps.getMaxToolRounds() * 2
+            }
+        });
+        it("uses request.maxToolRounds when it is explicitly defined", async () => {
+            const deps = makeDeps({ getMaxToolRounds: () => 50 });
+            const runner = new StreamRunner(deps);
+            const outcome = await runner.setup({
+                agent: "explore",
+                description: "test",
+                prompt: "test",
+                maxToolRounds: 75,
+            });
+            expect(outcome.kind).toBe("prepared");
+            if (outcome.kind === "prepared") {
+                expect(outcome.prepared.maxSteps).toBe(75);
+            }
+        });
+    });
 });
 //# sourceMappingURL=stream-runner.test.js.map

package/dist/src/orchestrator/agent.test.js

@@ -76,13 +76,16 @@ describe("Agent class", { timeout: 30_000 }, () => {
         agent.setModel("claude-sonnet-4-6-20250514");
         expect(agent.getModel()).toBe("claude-sonnet-4-6-20250514");
     });
-    it("constructs with sandbox mode", async () => {
+    it("constructs with sandbox mode — always returns 'off' (sandbox removed)", async () => {
         const { Agent } = await importAgentModule();
+        // Sandbox has been removed; passing sandboxMode: "shuru" is accepted for
+        // back-compat but getSandboxMode() always returns "off" until a new
+        // sandbox implementation is wired in.
         const agent = new Agent(undefined, undefined, undefined, undefined, {
             persistSession: false,
             sandboxMode: "shuru",
         });
-        expect(agent.getSandboxMode()).toBe("shuru");
+        expect(agent.getSandboxMode()).toBe("off");
     });
     it("defaults sandbox mode to off", async () => {
         const { Agent } = await importAgentModule();