@oxagen/cli 0.5.0 → 0.6.1

package/dist/agent/__tests__/git-isolation.test.js

@@ -0,0 +1,119 @@
+/**
+ * Git isolation protocol — proven against a *real* temp git repo, because the
+ * whole point is the actual git behavior, not a mock of it. We assert the four
+ * guarantees the fleet relies on:
+ *
+ *   1. No clobber — two agents editing the same file in their own worktrees
+ *      never see each other's writes.
+ *   2. Atomic + pinned — every checkpoint is one commit, pinned under
+ *      `refs/oxagen/agents/*` so it survives branch deletion and gc.
+ *   3. Conflicts surface, never corrupt — an overlapping merge is aborted, the
+ *      integration tree stays clean, and the losing branch is still recoverable.
+ *   4. Durable log — cleanup removes worktrees but keeps the pins.
+ */
+import { execFileSync } from "node:child_process";
+import { existsSync, mkdtempSync, rmSync, writeFileSync, readFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { afterEach, beforeEach, describe, expect, it } from "vitest";
+import { WorktreeManager } from "../fleet/git-isolation.js";
+function git(cwd, ...args) {
+    return execFileSync("git", args, { cwd, encoding: "utf8" }).trim();
+}
+let repo;
+beforeEach(() => {
+    repo = mkdtempSync(join(tmpdir(), "oxa-iso-"));
+    git(repo, "init", "-b", "main");
+    git(repo, "config", "user.email", "t@oxagen.dev");
+    git(repo, "config", "user.name", "Isolation Test");
+    git(repo, "config", "commit.gpgsign", "false");
+    writeFileSync(join(repo, "base.txt"), "original\n");
+    git(repo, "add", "-A");
+    git(repo, "commit", "-m", "init");
+});
+afterEach(() => {
+    rmSync(repo, { recursive: true, force: true });
+});
+describe("WorktreeManager", () => {
+    it("isolates parallel agents — same-file edits never clobber, both pinned", async () => {
+        const mgr = new WorktreeManager({ repoRoot: repo, namespace: "p1" });
+        const a = await mgr.spawn("task-a");
+        const b = await mgr.spawn("task-b");
+        expect(a).not.toBe(b);
+        // Both agents edit the SAME path — but in their own trees.
+        writeFileSync(join(a, "shared.txt"), "from A\n");
+        writeFileSync(join(b, "shared.txt"), "from B\n");
+        const cpA = await mgr.checkpoint("task-a", "a writes shared");
+        const cpB = await mgr.checkpoint("task-b", "b writes shared");
+        // No clobber: each worktree kept its own content.
+        expect(readFileSync(join(a, "shared.txt"), "utf8")).toBe("from A\n");
+        expect(readFileSync(join(b, "shared.txt"), "utf8")).toBe("from B\n");
+        // Atomic + pinned: one commit each, both reachable under refs/oxagen/agents.
+        expect(cpA?.hash).toMatch(/^[0-9a-f]{40}$/);
+        expect(cpB?.hash).toMatch(/^[0-9a-f]{40}$/);
+        expect(cpA.hash).not.toBe(cpB.hash);
+        const pins = git(repo, "for-each-ref", "--format=%(objectname)", "refs/oxagen/agents");
+        expect(pins).toContain(cpA.hash);
+        expect(pins).toContain(cpB.hash);
+    });
+    it("checkpoint returns null when the agent changed nothing", async () => {
+        const mgr = new WorktreeManager({ repoRoot: repo, namespace: "p2" });
+        await mgr.spawn("noop");
+        expect(await mgr.checkpoint("noop", "nothing")).toBeNull();
+    });
+    it("integrates non-overlapping work cleanly", async () => {
+        const mgr = new WorktreeManager({ repoRoot: repo, namespace: "p3" });
+        const a = await mgr.spawn("a");
+        writeFileSync(join(a, "a.txt"), "a\n");
+        await mgr.checkpoint("a", "add a.txt");
+        const rA = await mgr.integrate("a");
+        expect(rA.ok).toBe(true);
+        const b = await mgr.spawn("b");
+        writeFileSync(join(b, "b.txt"), "b\n");
+        await mgr.checkpoint("b", "add b.txt");
+        const rB = await mgr.integrate("b");
+        expect(rB.ok).toBe(true);
+        // The integration tree has both agents' files.
+        const int = mgr.integrationRef();
+        expect(existsSync(join(int.path, "a.txt"))).toBe(true);
+        expect(existsSync(join(int.path, "b.txt"))).toBe(true);
+    });
+    it("surfaces a conflict, aborts clean, and keeps the losing work recoverable", async () => {
+        const mgr = new WorktreeManager({ repoRoot: repo, namespace: "p4" });
+        // A and B both rewrite base.txt from the same frozen base → real conflict.
+        const a = await mgr.spawn("a");
+        writeFileSync(join(a, "base.txt"), "A version\n");
+        await mgr.checkpoint("a", "a edits base");
+        const rA = await mgr.integrate("a");
+        expect(rA.ok).toBe(true);
+        const intTipAfterA = git(repo, "rev-parse", mgr.integrationRef().branch);
+        const b = await mgr.spawn("b");
+        writeFileSync(join(b, "base.txt"), "B version\n");
+        const cpB = await mgr.checkpoint("b", "b edits base");
+        const rB = await mgr.integrate("b");
+        // Conflict surfaced, not silently merged.
+        expect(rB.ok).toBe(false);
+        expect(rB.conflicts).toContain("base.txt");
+        // Integration tree stayed clean (merge --abort) and branch is unchanged.
+        const int = mgr.integrationRef();
+        expect(git(int.path, "status", "--porcelain")).toBe("");
+        expect(git(repo, "rev-parse", int.branch)).toBe(intTipAfterA);
+        // The losing work is NOT lost — recoverable by its pinned tip.
+        expect(rB.taskTip).toBe(cpB.hash);
+        expect(git(repo, "cat-file", "-t", rB.taskTip)).toBe("commit");
+        expect(git(repo, "show", `${rB.taskTip}:base.txt`)).toBe("B version");
+    });
+    it("cleanupAll removes worktrees but keeps the durability pins", async () => {
+        const mgr = new WorktreeManager({ repoRoot: repo, namespace: "p5" });
+        const a = await mgr.spawn("a");
+        writeFileSync(join(a, "a.txt"), "a\n");
+        const cp = await mgr.checkpoint("a", "add a.txt");
+        await mgr.cleanupAll();
+        expect(existsSync(a)).toBe(false); // worktree gone
+        // Pin survives — the commit is still reachable by hash.
+        const pins = git(repo, "for-each-ref", "--format=%(objectname)", "refs/oxagen/agents");
+        expect(pins).toContain(cp.hash);
+        expect(git(repo, "cat-file", "-t", cp.hash)).toBe("commit");
+    });
+});
+//# sourceMappingURL=git-isolation.test.js.map

package/dist/agent/__tests__/git-isolation.test.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"git-isolation.test.js","sourceRoot":"","sources":["../../../src/agent/__tests__/git-isolation.test.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AACH,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,UAAU,EAAE,WAAW,EAAE,MAAM,EAAE,aAAa,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AACvF,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AACjC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,SAAS,EAAE,UAAU,EAAE,QAAQ,EAAE,MAAM,EAAE,EAAE,EAAE,MAAM,QAAQ,CAAC;AACrE,OAAO,EAAE,eAAe,EAAE,MAAM,2BAA2B,CAAC;AAE5D,SAAS,GAAG,CAAC,GAAW,EAAE,GAAG,IAAc;IACzC,OAAO,YAAY,CAAC,KAAK,EAAE,IAAI,EAAE,EAAE,GAAG,EAAE,QAAQ,EAAE,MAAM,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;AACrE,CAAC;AAED,IAAI,IAAY,CAAC;AAEjB,UAAU,CAAC,GAAG,EAAE;IACd,IAAI,GAAG,WAAW,CAAC,IAAI,CAAC,MAAM,EAAE,EAAE,UAAU,CAAC,CAAC,CAAC;IAC/C,GAAG,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,CAAC;IAChC,GAAG,CAAC,IAAI,EAAE,QAAQ,EAAE,YAAY,EAAE,cAAc,CAAC,CAAC;IAClD,GAAG,CAAC,IAAI,EAAE,QAAQ,EAAE,WAAW,EAAE,gBAAgB,CAAC,CAAC;IACnD,GAAG,CAAC,IAAI,EAAE,QAAQ,EAAE,gBAAgB,EAAE,OAAO,CAAC,CAAC;IAC/C,aAAa,CAAC,IAAI,CAAC,IAAI,EAAE,UAAU,CAAC,EAAE,YAAY,CAAC,CAAC;IACpD,GAAG,CAAC,IAAI,EAAE,KAAK,EAAE,IAAI,CAAC,CAAC;IACvB,GAAG,CAAC,IAAI,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,CAAC,CAAC;AACpC,CAAC,CAAC,CAAC;AAEH,SAAS,CAAC,GAAG,EAAE;IACb,MAAM,CAAC,IAAI,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;AACjD,CAAC,CAAC,CAAC;AAEH,QAAQ,CAAC,iBAAiB,EAAE,GAAG,EAAE;IAC/B,EAAE,CAAC,uEAAuE,EAAE,KAAK,IAAI,EAAE;QACrF,MAAM,GAAG,GAAG,IAAI,eAAe,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QACrE,MAAM,CAAC,GAAG,MAAM,GAAG,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;QACpC,MAAM,CAAC,GAAG,MAAM,GAAG,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC;QACpC,MAAM,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAEtB,2DAA2D;QAC3D,aAAa,CAAC,IAAI,CAAC,CAAC,EAAE,YAAY,CAAC,EAAE,UAAU,CAAC,CAAC;QACjD,aAAa,CAAC,IAAI,CAAC,CAAC,EAAE,YAAY,CAAC,EAAE,UAAU,CAAC,CAAC;QAEjD,MAAM,GAAG,GAAG,MAAM,GAAG,CAAC,UAAU,CAAC,QAAQ,EAAE,iBAAiB,CAAC,CAAC;QAC9D,MAAM,GAAG,GAAG,MAAM,GAAG,CAAC,UAAU,CAAC,QAAQ,EAAE,iBAAiB,CAAC,CAAC;QAE9D,kDAAkD;QAClD,MAAM,CAAC,YAAY,CAAC,IAAI,CAAC,CAAC,EAAE,YAAY,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;QACrE,MAAM,CAAC,YAAY,CAAC,IAAI,CAAC,CAAC,EAAE,YAAY,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;QAErE,6EAA6E;QAC7E,MAAM,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC,OAAO,CAAC,gBAAgB,CAAC,CAAC;QAC5C,MAAM,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC,OAAO,CAAC,gBAAgB,CAAC,CAAC;QAC5C,MAAM,CAAC,GAAI,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,GAAI,CAAC,IAAI,CAAC,CAAC;QACtC,MAAM,IAAI,GAAG,GAAG,CAAC,IAAI,EAAE,cAAc,EAAE,wBAAwB,EAAE,oBAAoB,CAAC,CAAC;QACvF,MAAM,CAAC,IAAI,CAAC,CAAC,SAAS,CAAC,GAAI,CAAC,IAAI,CAAC,CAAC;QAClC,MAAM,CAAC,IAAI,CAAC,CAAC,SAAS,CAAC,GAAI,CAAC,IAAI,CAAC,CAAC;IACpC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,wDAAwD,EAAE,KAAK,IAAI,EAAE;QACtE,MAAM,GAAG,GAAG,IAAI,eAAe,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QACrE,MAAM,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;QACxB,MAAM,CAAC,MAAM,GAAG,CAAC,UAAU,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC,CAAC,QAAQ,EAAE,CAAC;IAC7D,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,yCAAyC,EAAE,KAAK,IAAI,EAAE;QACvD,MAAM,GAAG,GAAG,IAAI,eAAe,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QAErE,MAAM,CAAC,GAAG,MAAM,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QAC/B,aAAa,CAAC,IAAI,CAAC,CAAC,EAAE,OAAO,CAAC,EAAE,KAAK,CAAC,CAAC;QACvC,MAAM,GAAG,CAAC,UAAU,CAAC,GAAG,EAAE,WAAW,CAAC,CAAC;QACvC,MAAM,EAAE,GAAG,MAAM,GAAG,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC;QACpC,MAAM,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAEzB,MAAM,CAAC,GAAG,MAAM,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QAC/B,aAAa,CAAC,IAAI,CAAC,CAAC,EAAE,OAAO,CAAC,EAAE,KAAK,CAAC,CAAC;QACvC,MAAM,GAAG,CAAC,UAAU,CAAC,GAAG,EAAE,WAAW,CAAC,CAAC;QACvC,MAAM,EAAE,GAAG,MAAM,GAAG,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC;QACpC,MAAM,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAEzB,+CAA+C;QAC/C,MAAM,GAAG,GAAG,GAAG,CAAC,cAAc,EAAG,CAAC;QAClC,MAAM,CAAC,UAAU,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACvD,MAAM,CAAC,UAAU,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACzD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,0EAA0E,EAAE,KAAK,IAAI,EAAE;QACxF,MAAM,GAAG,GAAG,IAAI,eAAe,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QAErE,2EAA2E;QAC3E,MAAM,CAAC,GAAG,MAAM,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QAC/B,aAAa,CAAC,IAAI,CAAC,CAAC,EAAE,UAAU,CAAC,EAAE,aAAa,CAAC,CAAC;QAClD,MAAM,GAAG,CAAC,UAAU,CAAC,GAAG,EAAE,cAAc,CAAC,CAAC;QAC1C,MAAM,EAAE,GAAG,MAAM,GAAG,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC;QACpC,MAAM,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACzB,MAAM,YAAY,GAAG,GAAG,CAAC,IAAI,EAAE,WAAW,EAAE,GAAG,CAAC,cAAc,EAAG,CAAC,MAAM,CAAC,CAAC;QAE1E,MAAM,CAAC,GAAG,MAAM,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QAC/B,aAAa,CAAC,IAAI,CAAC,CAAC,EAAE,UAAU,CAAC,EAAE,aAAa,CAAC,CAAC;QAClD,MAAM,GAAG,GAAG,MAAM,GAAG,CAAC,UAAU,CAAC,GAAG,EAAE,cAAc,CAAC,CAAC;QACtD,MAAM,EAAE,GAAG,MAAM,GAAG,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC;QAEpC,0CAA0C;QAC1C,MAAM,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QAC1B,MAAM,CAAC,EAAE,CAAC,SAAS,CAAC,CAAC,SAAS,CAAC,UAAU,CAAC,CAAC;QAE3C,yEAAyE;QACzE,MAAM,GAAG,GAAG,GAAG,CAAC,cAAc,EAAG,CAAC;QAClC,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,EAAE,QAAQ,EAAE,aAAa,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACxD,MAAM,CAAC,GAAG,CAAC,IAAI,EAAE,WAAW,EAAE,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;QAE9D,+DAA+D;QAC/D,MAAM,CAAC,EAAE,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,GAAI,CAAC,IAAI,CAAC,CAAC;QACnC,MAAM,CAAC,GAAG,CAAC,IAAI,EAAE,UAAU,EAAE,IAAI,EAAE,EAAE,CAAC,OAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QAChE,MAAM,CAAC,GAAG,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,EAAE,CAAC,OAAO,WAAW,CAAC,CAAC,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;IACxE,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,4DAA4D,EAAE,KAAK,IAAI,EAAE;QAC1E,MAAM,GAAG,GAAG,IAAI,eAAe,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QACrE,MAAM,CAAC,GAAG,MAAM,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QAC/B,aAAa,CAAC,IAAI,CAAC,CAAC,EAAE,OAAO,CAAC,EAAE,KAAK,CAAC,CAAC;QACvC,MAAM,EAAE,GAAG,MAAM,GAAG,CAAC,UAAU,CAAC,GAAG,EAAE,WAAW,CAAC,CAAC;QAElD,MAAM,GAAG,CAAC,UAAU,EAAE,CAAC;QAEvB,MAAM,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,gBAAgB;QACnD,wDAAwD;QACxD,MAAM,IAAI,GAAG,GAAG,CAAC,IAAI,EAAE,cAAc,EAAE,wBAAwB,EAAE,oBAAoB,CAAC,CAAC;QACvF,MAAM,CAAC,IAAI,CAAC,CAAC,SAAS,CAAC,EAAG,CAAC,IAAI,CAAC,CAAC;QACjC,MAAM,CAAC,GAAG,CAAC,IAAI,EAAE,UAAU,EAAE,IAAI,EAAE,EAAG,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IAC/D,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}

package/dist/agent/__tests__/judge.test.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=judge.test.d.ts.map

package/dist/agent/__tests__/judge.test.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"judge.test.d.ts","sourceRoot":"","sources":["../../../src/agent/__tests__/judge.test.ts"],"names":[],"mappings":""}

package/dist/agent/__tests__/judge.test.js

@@ -0,0 +1,140 @@
+/**
+ * Completeness judge — proves the advisor is always a DIFFERENT model than the
+ * executor, that a model verdict is returned faithfully, that the heuristic
+ * fallback flags "claimed a change but touched nothing", and that the revision
+ * prompt is built from the verdict. The model call is mocked.
+ */
+import { describe, it, expect, beforeEach, vi } from "vitest";
+vi.mock("ai", () => ({ generateObject: vi.fn() }));
+vi.mock("../../lib/config.js", () => ({ readConfig: () => ({}) }));
+import { generateObject } from "ai";
+import { judgeCompleteness, pickAdvisorModel, buildRevisionPrompt } from "../judge.js";
+const mockGen = generateObject;
+beforeEach(() => {
+    delete process.env["OXAGEN_LLM_ADVISOR"];
+    delete process.env["OXAGEN_LLM_FAST"];
+    delete process.env["OXAGEN_LLM_BALANCED"];
+    delete process.env["OXAGEN_LLM_PRECISE"];
+});
+const baseJudge = {
+    request: "add a route",
+    response: "Done, added the route.",
+    filesTouched: ["src/route.ts"],
+    commandsRun: [],
+    steps: 3,
+};
+describe("pickAdvisorModel", () => {
+    it("defaults to the most powerful OpenAI model (cross-vendor independence)", () => {
+        expect(pickAdvisorModel("anthropic/claude-sonnet-4.6")).toBe("openai/gpt-5.5-pro");
+    });
+    it("stays the OpenAI advisor even when the executor is the precise (Opus) tier", () => {
+        // A different vendor shares none of the executor's blind spots, so an Opus
+        // executor still gets the OpenAI judge — no need to fall back.
+        expect(pickAdvisorModel("anthropic/claude-opus-4.8")).toBe("openai/gpt-5.5-pro");
+    });
+    it("falls back to a distinct strong model when the executor IS the default advisor", () => {
+        const advisor = pickAdvisorModel("openai/gpt-5.5-pro");
+        expect(advisor).not.toBe("openai/gpt-5.5-pro");
+        expect(advisor).toContain("opus");
+    });
+    it("honours an explicit override that differs from the executor", () => {
+        process.env["OXAGEN_LLM_ADVISOR"] = "openai/gpt-5";
+        expect(pickAdvisorModel("anthropic/claude-sonnet-4.6")).toBe("openai/gpt-5");
+    });
+    it("ignores an override equal to the executor (must stay distinct)", () => {
+        process.env["OXAGEN_LLM_ADVISOR"] = "anthropic/claude-sonnet-4.6";
+        const advisor = pickAdvisorModel("anthropic/claude-sonnet-4.6");
+        expect(advisor).not.toBe("anthropic/claude-sonnet-4.6");
+    });
+});
+describe("judgeCompleteness", () => {
+    it("returns the advisor's verdict, judged by a different model", async () => {
+        mockGen.mockResolvedValueOnce({
+            object: {
+                complete: false,
+                confidence: 90,
+                findings: ["no test added"],
+                remainingWork: ["add a test"],
+                reasoning: "tests were promised but absent",
+            },
+            usage: { inputTokens: 4, outputTokens: 2 },
+        });
+        const v = await judgeCompleteness({ ...baseJudge, executorModel: "anthropic/claude-sonnet-4.6" });
+        expect(v.complete).toBe(false);
+        expect(v.findings).toEqual(["no test added"]);
+        expect(v.model).not.toBe("anthropic/claude-sonnet-4.6");
+        expect(v.fallback).toBe(false);
+        expect(v.usage.inputTokens).toBe(4);
+    });
+    it("clamps confidence into range", async () => {
+        mockGen.mockResolvedValueOnce({
+            object: { complete: true, confidence: 200, findings: [], remainingWork: [], reasoning: "ok" },
+            usage: {},
+        });
+        const v = await judgeCompleteness({ ...baseJudge, executorModel: "anthropic/claude-haiku-4.5" });
+        expect(v.confidence).toBe(100);
+    });
+    it("heuristic flags claimed-change-with-no-activity when the advisor is unavailable", async () => {
+        mockGen.mockRejectedValueOnce(new Error("down"));
+        const v = await judgeCompleteness({
+            request: "add a route",
+            response: "I added the route and updated the tests.",
+            filesTouched: [],
+            commandsRun: [],
+            steps: 1,
+            executorModel: "anthropic/claude-sonnet-4.6",
+        });
+        expect(v.fallback).toBe(true);
+        expect(v.complete).toBe(false);
+        expect(v.findings.length).toBeGreaterThan(0);
+    });
+    it("heuristic abstains to low-confidence complete when there is no red flag", async () => {
+        mockGen.mockRejectedValueOnce(new Error("down"));
+        const v = await judgeCompleteness({
+            request: "explain how auth works",
+            response: "Auth uses sessions stored in Postgres.",
+            filesTouched: [],
+            commandsRun: [],
+            steps: 1,
+            executorModel: "anthropic/claude-sonnet-4.6",
+        });
+        expect(v.fallback).toBe(true);
+        expect(v.complete).toBe(true);
+        expect(v.confidence).toBe(30);
+    });
+});
+describe("buildRevisionPrompt", () => {
+    it("renders findings and remaining work into actionable instructions", () => {
+        const verdict = {
+            complete: false,
+            confidence: 80,
+            findings: ["missing migration"],
+            remainingWork: ["write the migration"],
+            reasoning: "r",
+            model: "anthropic/claude-opus-4.8",
+            fallback: false,
+            usage: { inputTokens: 0, outputTokens: 0, costUsd: 0 },
+        };
+        const p = buildRevisionPrompt(verdict);
+        expect(p).toContain("NOT done");
+        expect(p).toContain("missing migration");
+        expect(p).toContain("write the migration");
+    });
+    it("omits empty sections cleanly", () => {
+        const verdict = {
+            complete: false,
+            confidence: 50,
+            findings: [],
+            remainingWork: [],
+            reasoning: "",
+            model: "m",
+            fallback: false,
+            usage: { inputTokens: 0, outputTokens: 0, costUsd: 0 },
+        };
+        const p = buildRevisionPrompt(verdict);
+        expect(p).not.toContain("Gaps found:");
+        expect(p).not.toContain("Remaining work:");
+        expect(p).toContain("Make the actual changes now");
+    });
+});
+//# sourceMappingURL=judge.test.js.map

package/dist/agent/__tests__/judge.test.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"judge.test.js","sourceRoot":"","sources":["../../../src/agent/__tests__/judge.test.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AACH,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,UAAU,EAAE,EAAE,EAAa,MAAM,QAAQ,CAAC;AAEzE,EAAE,CAAC,IAAI,CAAC,IAAI,EAAE,GAAG,EAAE,CAAC,CAAC,EAAE,cAAc,EAAE,EAAE,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC,CAAC;AACnD,EAAE,CAAC,IAAI,CAAC,qBAAqB,EAAE,GAAG,EAAE,CAAC,CAAC,EAAE,UAAU,EAAE,GAAG,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC;AAEnE,OAAO,EAAE,cAAc,EAAE,MAAM,IAAI,CAAC;AACpC,OAAO,EAAE,iBAAiB,EAAE,gBAAgB,EAAE,mBAAmB,EAAE,MAAM,aAAa,CAAC;AAGvF,MAAM,OAAO,GAAG,cAAiC,CAAC;AAElD,UAAU,CAAC,GAAG,EAAE;IACd,OAAO,OAAO,CAAC,GAAG,CAAC,oBAAoB,CAAC,CAAC;IACzC,OAAO,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC,CAAC;IACtC,OAAO,OAAO,CAAC,GAAG,CAAC,qBAAqB,CAAC,CAAC;IAC1C,OAAO,OAAO,CAAC,GAAG,CAAC,oBAAoB,CAAC,CAAC;AAC3C,CAAC,CAAC,CAAC;AAEH,MAAM,SAAS,GAAG;IAChB,OAAO,EAAE,aAAa;IACtB,QAAQ,EAAE,wBAAwB;IAClC,YAAY,EAAE,CAAC,cAAc,CAAC;IAC9B,WAAW,EAAE,EAAE;IACf,KAAK,EAAE,CAAC;CACT,CAAC;AAEF,QAAQ,CAAC,kBAAkB,EAAE,GAAG,EAAE;IAChC,EAAE,CAAC,wEAAwE,EAAE,GAAG,EAAE;QAChF,MAAM,CAAC,gBAAgB,CAAC,6BAA6B,CAAC,CAAC,CAAC,IAAI,CAAC,oBAAoB,CAAC,CAAC;IACrF,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,4EAA4E,EAAE,GAAG,EAAE;QACpF,2EAA2E;QAC3E,+DAA+D;QAC/D,MAAM,CAAC,gBAAgB,CAAC,2BAA2B,CAAC,CAAC,CAAC,IAAI,CAAC,oBAAoB,CAAC,CAAC;IACnF,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,gFAAgF,EAAE,GAAG,EAAE;QACxF,MAAM,OAAO,GAAG,gBAAgB,CAAC,oBAAoB,CAAC,CAAC;QACvD,MAAM,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,oBAAoB,CAAC,CAAC;QAC/C,MAAM,CAAC,OAAO,CAAC,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;IACpC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,6DAA6D,EAAE,GAAG,EAAE;QACrE,OAAO,CAAC,GAAG,CAAC,oBAAoB,CAAC,GAAG,cAAc,CAAC;QACnD,MAAM,CAAC,gBAAgB,CAAC,6BAA6B,CAAC,CAAC,CAAC,IAAI,CAAC,cAAc,CAAC,CAAC;IAC/E,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,gEAAgE,EAAE,GAAG,EAAE;QACxE,OAAO,CAAC,GAAG,CAAC,oBAAoB,CAAC,GAAG,6BAA6B,CAAC;QAClE,MAAM,OAAO,GAAG,gBAAgB,CAAC,6BAA6B,CAAC,CAAC;QAChE,MAAM,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,6BAA6B,CAAC,CAAC;IAC1D,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,QAAQ,CAAC,mBAAmB,EAAE,GAAG,EAAE;IACjC,EAAE,CAAC,4DAA4D,EAAE,KAAK,IAAI,EAAE;QAC1E,OAAO,CAAC,qBAAqB,CAAC;YAC5B,MAAM,EAAE;gBACN,QAAQ,EAAE,KAAK;gBACf,UAAU,EAAE,EAAE;gBACd,QAAQ,EAAE,CAAC,eAAe,CAAC;gBAC3B,aAAa,EAAE,CAAC,YAAY,CAAC;gBAC7B,SAAS,EAAE,gCAAgC;aAC5C;YACD,KAAK,EAAE,EAAE,WAAW,EAAE,CAAC,EAAE,YAAY,EAAE,CAAC,EAAE;SAC3C,CAAC,CAAC;QACH,MAAM,CAAC,GAAG,MAAM,iBAAiB,CAAC,EAAE,GAAG,SAAS,EAAE,aAAa,EAAE,6BAA6B,EAAE,CAAC,CAAC;QAClG,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QAC/B,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,OAAO,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC;QAC9C,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,6BAA6B,CAAC,CAAC;QACxD,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QAC/B,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IACtC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,8BAA8B,EAAE,KAAK,IAAI,EAAE;QAC5C,OAAO,CAAC,qBAAqB,CAAC;YAC5B,MAAM,EAAE,EAAE,QAAQ,EAAE,IAAI,EAAE,UAAU,EAAE,GAAG,EAAE,QAAQ,EAAE,EAAE,EAAE,aAAa,EAAE,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE;YAC7F,KAAK,EAAE,EAAE;SACV,CAAC,CAAC;QACH,MAAM,CAAC,GAAG,MAAM,iBAAiB,CAAC,EAAE,GAAG,SAAS,EAAE,aAAa,EAAE,4BAA4B,EAAE,CAAC,CAAC;QACjG,MAAM,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IACjC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,iFAAiF,EAAE,KAAK,IAAI,EAAE;QAC/F,OAAO,CAAC,qBAAqB,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC;QACjD,MAAM,CAAC,GAAG,MAAM,iBAAiB,CAAC;YAChC,OAAO,EAAE,aAAa;YACtB,QAAQ,EAAE,0CAA0C;YACpD,YAAY,EAAE,EAAE;YAChB,WAAW,EAAE,EAAE;YACf,KAAK,EAAE,CAAC;YACR,aAAa,EAAE,6BAA6B;SAC7C,CAAC,CAAC;QACH,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAC9B,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QAC/B,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC;IAC/C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,yEAAyE,EAAE,KAAK,IAAI,EAAE;QACvF,OAAO,CAAC,qBAAqB,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC;QACjD,MAAM,CAAC,GAAG,MAAM,iBAAiB,CAAC;YAChC,OAAO,EAAE,wBAAwB;YACjC,QAAQ,EAAE,wCAAwC;YAClD,YAAY,EAAE,EAAE;YAChB,WAAW,EAAE,EAAE;YACf,KAAK,EAAE,CAAC;YACR,aAAa,EAAE,6BAA6B;SAC7C,CAAC,CAAC;QACH,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAC9B,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAC9B,MAAM,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAChC,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,QAAQ,CAAC,qBAAqB,EAAE,GAAG,EAAE;IACnC,EAAE,CAAC,kEAAkE,EAAE,GAAG,EAAE;QAC1E,MAAM,OAAO,GAAiB;YAC5B,QAAQ,EAAE,KAAK;YACf,UAAU,EAAE,EAAE;YACd,QAAQ,EAAE,CAAC,mBAAmB,CAAC;YAC/B,aAAa,EAAE,CAAC,qBAAqB,CAAC;YACtC,SAAS,EAAE,GAAG;YACd,KAAK,EAAE,2BAA2B;YAClC,QAAQ,EAAE,KAAK;YACf,KAAK,EAAE,EAAE,WAAW,EAAE,CAAC,EAAE,YAAY,EAAE,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE;SACvD,CAAC;QACF,MAAM,CAAC,GAAG,mBAAmB,CAAC,OAAO,CAAC,CAAC;QACvC,MAAM,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,UAAU,CAAC,CAAC;QAChC,MAAM,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,mBAAmB,CAAC,CAAC;QACzC,MAAM,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,qBAAqB,CAAC,CAAC;IAC7C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,8BAA8B,EAAE,GAAG,EAAE;QACtC,MAAM,OAAO,GAAiB;YAC5B,QAAQ,EAAE,KAAK;YACf,UAAU,EAAE,EAAE;YACd,QAAQ,EAAE,EAAE;YACZ,aAAa,EAAE,EAAE;YACjB,SAAS,EAAE,EAAE;YACb,KAAK,EAAE,GAAG;YACV,QAAQ,EAAE,KAAK;YACf,KAAK,EAAE,EAAE,WAAW,EAAE,CAAC,EAAE,YAAY,EAAE,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE;SACvD,CAAC;QACF,MAAM,CAAC,GAAG,mBAAmB,CAAC,OAAO,CAAC,CAAC;QACvC,MAAM,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,SAAS,CAAC,aAAa,CAAC,CAAC;QACvC,MAAM,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,SAAS,CAAC,iBAAiB,CAAC,CAAC;QAC3C,MAAM,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,6BAA6B,CAAC,CAAC;IACrD,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}

package/dist/agent/__tests__/loop-errors.test.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=loop-errors.test.d.ts.map

package/dist/agent/__tests__/loop-errors.test.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"loop-errors.test.d.ts","sourceRoot":"","sources":["../../../src/agent/__tests__/loop-errors.test.ts"],"names":[],"mappings":""}

package/dist/agent/__tests__/loop-errors.test.js

@@ -0,0 +1,75 @@
+/**
+ * normalizeAgentError turns raw AI Gateway / stream errors into one clean,
+ * actionable line — so dogfooding failures (no credits, bad key, rate limit)
+ * are legible instead of dumping the SDK's internal error object.
+ */
+import { describe, it, expect } from "vitest";
+import { normalizeAgentError, isErrorResult, stringifyCapped } from "../loop.js";
+/** Shape of an AI SDK gateway error: provider JSON lives on `responseBody`. */
+function gatewayError(message, type = "error") {
+    const err = new Error("No output generated. Check the stream for errors.");
+    err.responseBody = JSON.stringify({
+        error: { message, type },
+    });
+    return err;
+}
+describe("normalizeAgentError", () => {
+    it("translates insufficient-funds into an add-credits message", () => {
+        const out = normalizeAgentError(gatewayError("A positive credit balance is required for all requests, including BYOK.", "insufficient_funds"));
+        expect(out.message).toContain("no credit balance");
+        expect(out.message).toContain("Vercel AI Gateway");
+        // The opaque SDK wrapper message must not leak through.
+        expect(out.message).not.toContain("No output generated");
+    });
+    it("translates a 401 into a key-check message", () => {
+        const out = normalizeAgentError(gatewayError("401 Unauthorized: invalid api key"));
+        expect(out.message).toContain("401");
+        expect(out.message).toContain("AI_GATEWAY_API_KEY");
+    });
+    it("translates a 429 into a rate-limit message", () => {
+        const out = normalizeAgentError(gatewayError("429 rate limit exceeded"));
+        expect(out.message).toContain("429");
+        expect(out.message.toLowerCase()).toContain("rate");
+    });
+    it("passes through an ordinary Error unchanged", () => {
+        const original = new Error("something specific broke");
+        expect(normalizeAgentError(original)).toBe(original);
+    });
+    it("wraps a non-Error value into an Error", () => {
+        const out = normalizeAgentError("plain string failure");
+        expect(out).toBeInstanceOf(Error);
+        expect(out.message).toContain("plain string failure");
+    });
+});
+describe("isErrorResult — sets the ok flag on tool events", () => {
+    it("flags an Error instance", () => {
+        expect(isErrorResult(new Error("boom"))).toBe(true);
+    });
+    it("flags an object with isError:true or a truthy error", () => {
+        expect(isErrorResult({ isError: true })).toBe(true);
+        expect(isErrorResult({ error: "ENOENT" })).toBe(true);
+    });
+    it("treats a normal result (incl. error:false / error:null) as ok", () => {
+        expect(isErrorResult({ ok: true, output: "done" })).toBe(false);
+        expect(isErrorResult({ error: false })).toBe(false);
+        expect(isErrorResult({ error: null })).toBe(false);
+        expect(isErrorResult("plain string output")).toBe(false);
+        expect(isErrorResult(undefined)).toBe(false);
+    });
+});
+describe("stringifyCapped — bounds captured tool input/results", () => {
+    it("passes a short string through and JSON-stringifies objects", () => {
+        expect(stringifyCapped("hi", 100)).toBe("hi");
+        expect(stringifyCapped({ path: "src/x.ts" }, 100)).toBe('{"path":"src/x.ts"}');
+    });
+    it("truncates with an ellipsis past the cap", () => {
+        const out = stringifyCapped("x".repeat(50), 10);
+        expect(out).toBe("x".repeat(10) + "…");
+    });
+    it("falls back to String() for unserializable values", () => {
+        const circular = {};
+        circular.self = circular;
+        expect(typeof stringifyCapped(circular, 100)).toBe("string");
+    });
+});
+//# sourceMappingURL=loop-errors.test.js.map

package/dist/agent/__tests__/loop-errors.test.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"loop-errors.test.js","sourceRoot":"","sources":["../../../src/agent/__tests__/loop-errors.test.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AACH,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,MAAM,QAAQ,CAAC;AAC9C,OAAO,EAAE,mBAAmB,EAAE,aAAa,EAAE,eAAe,EAAE,MAAM,YAAY,CAAC;AAEjF,+EAA+E;AAC/E,SAAS,YAAY,CAAC,OAAe,EAAE,IAAI,GAAG,OAAO;IACnD,MAAM,GAAG,GAAG,IAAI,KAAK,CAAC,mDAAmD,CAAC,CAAC;IAC1E,GAA2C,CAAC,YAAY,GAAG,IAAI,CAAC,SAAS,CAAC;QACzE,KAAK,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE;KACzB,CAAC,CAAC;IACH,OAAO,GAAG,CAAC;AACb,CAAC;AAED,QAAQ,CAAC,qBAAqB,EAAE,GAAG,EAAE;IACnC,EAAE,CAAC,2DAA2D,EAAE,GAAG,EAAE;QACnE,MAAM,GAAG,GAAG,mBAAmB,CAC7B,YAAY,CACV,yEAAyE,EACzE,oBAAoB,CACrB,CACF,CAAC;QACF,MAAM,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,SAAS,CAAC,mBAAmB,CAAC,CAAC;QACnD,MAAM,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,SAAS,CAAC,mBAAmB,CAAC,CAAC;QACnD,wDAAwD;QACxD,MAAM,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,SAAS,CAAC,qBAAqB,CAAC,CAAC;IAC3D,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,2CAA2C,EAAE,GAAG,EAAE;QACnD,MAAM,GAAG,GAAG,mBAAmB,CAAC,YAAY,CAAC,mCAAmC,CAAC,CAAC,CAAC;QACnF,MAAM,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;QACrC,MAAM,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,SAAS,CAAC,oBAAoB,CAAC,CAAC;IACtD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,4CAA4C,EAAE,GAAG,EAAE;QACpD,MAAM,GAAG,GAAG,mBAAmB,CAAC,YAAY,CAAC,yBAAyB,CAAC,CAAC,CAAC;QACzE,MAAM,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;QACrC,MAAM,CAAC,GAAG,CAAC,OAAO,CAAC,WAAW,EAAE,CAAC,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;IACtD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,4CAA4C,EAAE,GAAG,EAAE;QACpD,MAAM,QAAQ,GAAG,IAAI,KAAK,CAAC,0BAA0B,CAAC,CAAC;QACvD,MAAM,CAAC,mBAAmB,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IACvD,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,uCAAuC,EAAE,GAAG,EAAE;QAC/C,MAAM,GAAG,GAAG,mBAAmB,CAAC,sBAAsB,CAAC,CAAC;QACxD,MAAM,CAAC,GAAG,CAAC,CAAC,cAAc,CAAC,KAAK,CAAC,CAAC;QAClC,MAAM,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,SAAS,CAAC,sBAAsB,CAAC,CAAC;IACxD,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,QAAQ,CAAC,iDAAiD,EAAE,GAAG,EAAE;IAC/D,EAAE,CAAC,yBAAyB,EAAE,GAAG,EAAE;QACjC,MAAM,CAAC,aAAa,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACtD,CAAC,CAAC,CAAC;IACH,EAAE,CAAC,qDAAqD,EAAE,GAAG,EAAE;QAC7D,MAAM,CAAC,aAAa,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACpD,MAAM,CAAC,aAAa,CAAC,EAAE,KAAK,EAAE,QAAQ,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACxD,CAAC,CAAC,CAAC;IACH,EAAE,CAAC,+DAA+D,EAAE,GAAG,EAAE;QACvE,MAAM,CAAC,aAAa,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QAChE,MAAM,CAAC,aAAa,CAAC,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QACpD,MAAM,CAAC,aAAa,CAAC,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QACnD,MAAM,CAAC,aAAa,CAAC,qBAAqB,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QACzD,MAAM,CAAC,aAAa,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;IAC/C,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC;AAEH,QAAQ,CAAC,sDAAsD,EAAE,GAAG,EAAE;IACpE,EAAE,CAAC,4DAA4D,EAAE,GAAG,EAAE;QACpE,MAAM,CAAC,eAAe,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAC9C,MAAM,CAAC,eAAe,CAAC,EAAE,IAAI,EAAE,UAAU,EAAE,EAAE,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,qBAAqB,CAAC,CAAC;IACjF,CAAC,CAAC,CAAC;IACH,EAAE,CAAC,yCAAyC,EAAE,GAAG,EAAE;QACjD,MAAM,GAAG,GAAG,eAAe,CAAC,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC;QAChD,MAAM,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC;IACzC,CAAC,CAAC,CAAC;IACH,EAAE,CAAC,kDAAkD,EAAE,GAAG,EAAE;QAC1D,MAAM,QAAQ,GAA4B,EAAE,CAAC;QAC7C,QAAQ,CAAC,IAAI,GAAG,QAAQ,CAAC;QACzB,MAAM,CAAC,OAAO,eAAe,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IAC/D,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}

package/dist/agent/__tests__/loop-gating.test.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=loop-gating.test.d.ts.map

package/dist/agent/__tests__/loop-gating.test.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"loop-gating.test.d.ts","sourceRoot":"","sources":["../../../src/agent/__tests__/loop-gating.test.ts"],"names":[],"mappings":""}

package/dist/agent/__tests__/loop-gating.test.js

@@ -0,0 +1,94 @@
+/**
+ * Regression guard for a silent permission-gate bypass in runAgent.
+ *
+ * The loop builds a permission-gated, hook-wrapped tool set
+ * (`wrapToolsWithGate(buildTools(…, { broker }))`) and must hand THAT set to
+ * `streamText`. A merge once left `streamText` calling `buildTools` a second
+ * time — raw and ungated — so settings.permissions and Pre/PostToolUse hooks
+ * were silently skipped on every CLI agent turn. These tests pin the wiring:
+ *   1. `buildTools` is called exactly once, and with the broker threaded in.
+ *   2. the tools `streamText` receives are the gated ones (a denied tool's
+ *      execute returns the gate's deny string and never runs the real tool).
+ */
+import { describe, it, expect, vi, beforeEach } from "vitest";
+// Hoisted so the vi.mock factories (themselves hoisted above imports) can close
+// over initialised values instead of TDZ-dead consts.
+const h = vi.hoisted(() => {
+    const realExecute = vi.fn(async () => "WROTE FILE");
+    return {
+        streamTextMock: vi.fn(),
+        realExecute,
+        buildToolsMock: vi.fn((_cwd, _opts) => ({
+            write: {
+                description: "write a file",
+                inputSchema: { jsonSchema: { type: "object" } },
+                execute: realExecute,
+            },
+        })),
+    };
+});
+vi.mock("ai", async (importOriginal) => {
+    const actual = await importOriginal();
+    return { ...actual, streamText: (args) => h.streamTextMock(args) };
+});
+vi.mock("../tools.js", () => ({ buildTools: h.buildToolsMock }));
+// Keep the loop off the network / filesystem; leave the gate (real) intact.
+vi.mock("../env.js", () => ({
+    ensureGatewayKey: () => true,
+    MissingGatewayKeyError: class extends Error {
+    },
+}));
+vi.mock("../model.js", () => ({ resolveModelId: () => "test/model" }));
+vi.mock("../system-prompt.js", () => ({ buildSystemPrompt: () => "system" }));
+vi.mock("../../settings/hooks.js", () => ({
+    runHooks: vi.fn(async () => ({ blocked: false, output: "", reason: undefined })),
+}));
+import { runAgent } from "../loop.js";
+function fakeStream() {
+    return {
+        textStream: (async function* () {
+            yield "ok";
+        })(),
+        steps: Promise.resolve([]),
+        usage: Promise.resolve({ inputTokens: 1, outputTokens: 1, totalTokens: 2 }),
+        response: Promise.resolve({ messages: [] }),
+    };
+}
+describe("runAgent — tools reach streamText fully gated", () => {
+    beforeEach(() => {
+        h.streamTextMock.mockReset();
+        h.streamTextMock.mockImplementation(() => fakeStream());
+        h.buildToolsMock.mockClear();
+        h.realExecute.mockClear();
+    });
+    it("builds tools exactly once and threads the permission broker", async () => {
+        const broker = { id: "test-broker" };
+        await runAgent({
+            prompt: "do it",
+            broker,
+            memory: null,
+            settings: { permissions: {}, hooks: {} },
+        });
+        expect(h.buildToolsMock).toHaveBeenCalledTimes(1);
+        expect(h.buildToolsMock.mock.calls[0]?.[1]).toMatchObject({ broker });
+    });
+    it("hands streamText the gated tool set — a denied tool never runs", async () => {
+        await runAgent({
+            prompt: "write a file",
+            memory: null,
+            // Deny the write tool via settings permissions; the real gate enforces it.
+            settings: { permissions: { deny: ["write"] }, hooks: {} },
+        });
+        expect(h.streamTextMock).toHaveBeenCalledTimes(1);
+        const passed = h.streamTextMock.mock.calls[0]?.[0];
+        const gatedWrite = passed.tools.write;
+        if (!gatedWrite)
+            throw new Error("expected a gated `write` tool on streamText");
+        // Invoke the gated execute: it must short-circuit with a deny string and
+        // must NOT reach the underlying tool implementation.
+        const out = await gatedWrite.execute?.({}, {});
+        expect(String(out)).toContain("Permission denied");
+        expect(h.realExecute).not.toHaveBeenCalled();
+    });
+});
+//# sourceMappingURL=loop-gating.test.js.map

package/dist/agent/__tests__/loop-gating.test.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"loop-gating.test.js","sourceRoot":"","sources":["../../../src/agent/__tests__/loop-gating.test.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AACH,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE,UAAU,EAAE,MAAM,QAAQ,CAAC;AAG9D,gFAAgF;AAChF,sDAAsD;AACtD,MAAM,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,GAAG,EAAE;IACxB,MAAM,WAAW,GAAG,EAAE,CAAC,EAAE,CAAC,KAAK,IAAI,EAAE,CAAC,YAAY,CAAC,CAAC;IACpD,OAAO;QACL,cAAc,EAAE,EAAE,CAAC,EAAE,EAAE;QACvB,WAAW;QACX,cAAc,EAAE,EAAE,CAAC,EAAE,CACnB,CAAC,IAAY,EAAE,KAAgD,EAAE,EAAE,CAAC,CAAC;YACnE,KAAK,EAAE;gBACL,WAAW,EAAE,cAAc;gBAC3B,WAAW,EAAE,EAAE,UAAU,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE;gBAC/C,OAAO,EAAE,WAAW;aACrB;SACF,CAAC,CACH;KACF,CAAC;AACJ,CAAC,CAAC,CAAC;AAEH,EAAE,CAAC,IAAI,CAAC,IAAI,EAAE,KAAK,EAAE,cAAc,EAAE,EAAE;IACrC,MAAM,MAAM,GAAG,MAAM,cAAc,EAAuB,CAAC;IAC3D,OAAO,EAAE,GAAG,MAAM,EAAE,UAAU,EAAE,CAAC,IAAa,EAAE,EAAE,CAAC,CAAC,CAAC,cAAc,CAAC,IAAI,CAAC,EAAE,CAAC;AAC9E,CAAC,CAAC,CAAC;AACH,EAAE,CAAC,IAAI,CAAC,aAAa,EAAE,GAAG,EAAE,CAAC,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,cAAc,EAAE,CAAC,CAAC,CAAC;AAEjE,4EAA4E;AAC5E,EAAE,CAAC,IAAI,CAAC,WAAW,EAAE,GAAG,EAAE,CAAC,CAAC;IAC1B,gBAAgB,EAAE,GAAG,EAAE,CAAC,IAAI;IAC5B,sBAAsB,EAAE,KAAM,SAAQ,KAAK;KAAG;CAC/C,CAAC,CAAC,CAAC;AACJ,EAAE,CAAC,IAAI,CAAC,aAAa,EAAE,GAAG,EAAE,CAAC,CAAC,EAAE,cAAc,EAAE,GAAG,EAAE,CAAC,YAAY,EAAE,CAAC,CAAC,CAAC;AACvE,EAAE,CAAC,IAAI,CAAC,qBAAqB,EAAE,GAAG,EAAE,CAAC,CAAC,EAAE,iBAAiB,EAAE,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC;AAC9E,EAAE,CAAC,IAAI,CAAC,yBAAyB,EAAE,GAAG,EAAE,CAAC,CAAC;IACxC,QAAQ,EAAE,EAAE,CAAC,EAAE,CAAC,KAAK,IAAI,EAAE,CAAC,CAAC,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,EAAE,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;CACjF,CAAC,CAAC,CAAC;AAEJ,OAAO,EAAE,QAAQ,EAAE,MAAM,YAAY,CAAC;AAEtC,SAAS,UAAU;IACjB,OAAO;QACL,UAAU,EAAE,CAAC,KAAK,SAAS,CAAC;YAC1B,MAAM,IAAI,CAAC;QACb,CAAC,CAAC,EAAE;QACJ,KAAK,EAAE,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;QAC1B,KAAK,EAAE,OAAO,CAAC,OAAO,CAAC,EAAE,WAAW,EAAE,CAAC,EAAE,YAAY,EAAE,CAAC,EAAE,WAAW,EAAE,CAAC,EAAE,CAAC;QAC3E,QAAQ,EAAE,OAAO,CAAC,OAAO,CAAC,EAAE,QAAQ,EAAE,EAAE,EAAE,CAAC;KAC5C,CAAC;AACJ,CAAC;AAED,QAAQ,CAAC,+CAA+C,EAAE,GAAG,EAAE;IAC7D,UAAU,CAAC,GAAG,EAAE;QACd,CAAC,CAAC,cAAc,CAAC,SAAS,EAAE,CAAC;QAC7B,CAAC,CAAC,cAAc,CAAC,kBAAkB,CAAC,GAAG,EAAE,CAAC,UAAU,EAAE,CAAC,CAAC;QACxD,CAAC,CAAC,cAAc,CAAC,SAAS,EAAE,CAAC;QAC7B,CAAC,CAAC,WAAW,CAAC,SAAS,EAAE,CAAC;IAC5B,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,6DAA6D,EAAE,KAAK,IAAI,EAAE;QAC3E,MAAM,MAAM,GAAG,EAAE,EAAE,EAAE,aAAa,EAAW,CAAC;QAC9C,MAAM,QAAQ,CAAC;YACb,MAAM,EAAE,OAAO;YACf,MAAM;YACN,MAAM,EAAE,IAAI;YACZ,QAAQ,EAAE,EAAE,WAAW,EAAE,EAAE,EAAE,KAAK,EAAE,EAAE,EAAW;SAClD,CAAC,CAAC;QAEH,MAAM,CAAC,CAAC,CAAC,cAAc,CAAC,CAAC,qBAAqB,CAAC,CAAC,CAAC,CAAC;QAClD,MAAM,CAAC,CAAC,CAAC,cAAc,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC;IACxE,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,gEAAgE,EAAE,KAAK,IAAI,EAAE;QAC9E,MAAM,QAAQ,CAAC;YACb,MAAM,EAAE,cAAc;YACtB,MAAM,EAAE,IAAI;YACZ,2EAA2E;YAC3E,QAAQ,EAAE,EAAE,WAAW,EAAE,EAAE,IAAI,EAAE,CAAC,OAAO,CAAC,EAAE,EAAE,KAAK,EAAE,EAAE,EAAW;SACnE,CAAC,CAAC;QAEH,MAAM,CAAC,CAAC,CAAC,cAAc,CAAC,CAAC,qBAAqB,CAAC,CAAC,CAAC,CAAC;QAClD,MAAM,MAAM,GAAG,CAAC,CAAC,cAAc,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAuB,CAAC;QACzE,MAAM,UAAU,GAAG,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC;QACtC,IAAI,CAAC,UAAU;YAAE,MAAM,IAAI,KAAK,CAAC,6CAA6C,CAAC,CAAC;QAEhF,yEAAyE;QACzE,qDAAqD;QACrD,MAAM,GAAG,GAAG,MAAM,UAAU,CAAC,OAAO,EAAE,CAAC,EAAW,EAAE,EAAW,CAAC,CAAC;QACjE,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,SAAS,CAAC,mBAAmB,CAAC,CAAC;QACnD,MAAM,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,GAAG,CAAC,gBAAgB,EAAE,CAAC;IAC/C,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}

package/dist/agent/__tests__/model-router.test.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=model-router.test.d.ts.map

package/dist/agent/__tests__/model-router.test.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"model-router.test.d.ts","sourceRoot":"","sources":["../../../src/agent/__tests__/model-router.test.ts"],"names":[],"mappings":""}