@opengsd/gsd-pi 1.1.1-dev.74e8dd1 → 1.1.1-dev.a5a2de8

package/src/resources/extensions/gsd/tests/gate-1b-recovery-bound-corrections.test.ts

@@ -1,246 +0,0 @@
-// gsd-pi + Gate 1b recovery bound corrections — regression tests for the two bugs
-// found in peer review of the H1 fix (commit f0e1d42a2):
-//   1. Escalation message must describe /gsd (counter reset) AND /gsd-debug (diagnose).
-//   2. planBlockedRecoveryCount must NOT increment when pi.sendMessage throws.
-
-import { describe, test, beforeEach, afterEach } from "node:test";
-import assert from "node:assert/strict";
-import { mkdtempSync, mkdirSync, rmSync, writeFileSync } from "node:fs";
-import { join } from "node:path";
-import { tmpdir } from "node:os";
-
-import {
-  checkAutoStartAfterDiscuss,
-  setPendingAutoStart,
-  clearPendingAutoStart,
-  _getPendingAutoStart,
-} from "../guided-flow.ts";
-import { drainLogs } from "../workflow-logger.ts";
-import {
-  openDatabase,
-  closeDatabase,
-  insertMilestone,
-} from "../gsd-db.ts";
-
-// ─── Harness ───────────────────────────────────────────────────────────────
-
-interface MockCapture {
-  notifies: Array<{ msg: string; level: string }>;
-  messages: Array<{ payload: any; options: any }>;
-}
-
-function mkCapture(): MockCapture {
-  return { notifies: [], messages: [] };
-}
-
-function mkCtx(cap: MockCapture): any {
-  return {
-    ui: {
-      notify: (msg: string, level: string) => {
-        cap.notifies.push({ msg, level });
-      },
-    },
-  };
-}
-
-/** Returns a pi stub whose sendMessage throws on the first call, succeeds after. */
-function mkPiThrowOnce(cap: MockCapture): any {
-  let callCount = 0;
-  return {
-    sendMessage: (payload: any, options: any) => {
-      callCount += 1;
-      if (callCount === 1) {
-        throw new Error("transient network error");
-      }
-      cap.messages.push({ payload, options });
-    },
-    setActiveTools: () => undefined,
-    getActiveTools: () => [],
-  };
-}
-
-function mkPi(cap: MockCapture): any {
-  return {
-    sendMessage: (payload: any, options: any) => {
-      cap.messages.push({ payload, options });
-    },
-    setActiveTools: () => undefined,
-    getActiveTools: () => [],
-  };
-}
-
-function mkBase(): string {
-  const base = mkdtempSync(join(tmpdir(), "gsd-gate1b-corrections-"));
-  mkdirSync(join(base, ".gsd", "milestones", "M001"), { recursive: true });
-  writeFileSync(
-    join(base, ".gsd", "milestones", "M001", "M001-CONTEXT.md"),
-    "# M001: Corrections Test\n\nContext written by discuss phase.\n",
-  );
-  return base;
-}
-
-// ─── Tests ─────────────────────────────────────────────────────────────────
-
-describe("Gate 1b recovery bound corrections", () => {
-  let base: string;
-  let cap: MockCapture;
-
-  beforeEach(() => {
-    clearPendingAutoStart();
-    drainLogs();
-  });
-
-  afterEach(() => {
-    closeDatabase();
-    clearPendingAutoStart();
-    if (base) {
-      rmSync(base, { recursive: true, force: true });
-    }
-  });
-
-  // ── Fix 1: escalation message ──────────────────────────────────────────
-
-  test("escalation message describes /gsd for reset AND /gsd-debug for diagnosis", () => {
-    base = mkBase();
-    openDatabase(":memory:");
-    insertMilestone({ id: "M001", title: "Corrections Test", status: "queued" });
-
-    cap = mkCapture();
-    setPendingAutoStart(base, {
-      basePath: base,
-      milestoneId: "M001",
-      ctx: mkCtx(cap),
-      pi: mkPi(cap),
-    });
-
-    // Exhaust the recovery budget (MAX = 3)
-    checkAutoStartAfterDiscuss(); // count → 1
-    checkAutoStartAfterDiscuss(); // count → 2
-    checkAutoStartAfterDiscuss(); // count → 3
-
-    cap.notifies = [];
-    drainLogs();
-
-    // This call hits the cap and must escalate
-    const result = checkAutoStartAfterDiscuss();
-    assert.equal(result, false, "escalation call must return false");
-
-    const errorNotify = cap.notifies.find((n) => n.level === "error");
-    assert.ok(errorNotify, "escalation must emit a notify with level 'error'");
-
-    // Must mention /gsd with reset semantics
-    assert.match(
-      errorNotify.msg,
-      /\/gsd\b/,
-      "escalation message must reference /gsd (the command that resets the counter)",
-    );
-    assert.match(
-      errorNotify.msg,
-      /reset/i,
-      "escalation message must use the word 'reset' so users know /gsd resets the counter",
-    );
-
-    // Must also mention /gsd-debug
-    assert.match(
-      errorNotify.msg,
-      /\/gsd-debug/i,
-      "escalation message must also reference /gsd-debug for diagnosis",
-    );
-
-    // Must NOT suggest /gsd-debug alone as the sole remediation
-    assert.doesNotMatch(
-      errorNotify.msg,
-      /^[^/]*\/gsd-debug[^/]*$/,
-      "escalation message must not mention /gsd-debug as the only option",
-    );
-  });
-
-  // ── Fix 2: counter ordering ────────────────────────────────────────────
-
-  test("counter stays at 0 when sendMessage throws on the first call", () => {
-    base = mkBase();
-    openDatabase(":memory:");
-    insertMilestone({ id: "M001", title: "Corrections Test", status: "queued" });
-
-    cap = mkCapture();
-    setPendingAutoStart(base, {
-      basePath: base,
-      milestoneId: "M001",
-      ctx: mkCtx(cap),
-      pi: mkPiThrowOnce(cap),
-    });
-
-    // First call: sendMessage throws — counter must NOT increment
-    const result = checkAutoStartAfterDiscuss();
-    assert.equal(result, false, "must return false even when sendMessage throws");
-
-    const entry = _getPendingAutoStart(base);
-    assert.ok(entry, "entry must still exist after a failed sendMessage");
-    assert.equal(
-      entry.planBlockedRecoveryCount,
-      0,
-      "counter must remain 0 when sendMessage throws — no budget burned by transient failure",
-    );
-  });
-
-  test("counter increments to 1 on the second call when first sendMessage threw", () => {
-    base = mkBase();
-    openDatabase(":memory:");
-    insertMilestone({ id: "M001", title: "Corrections Test", status: "queued" });
-
-    cap = mkCapture();
-    setPendingAutoStart(base, {
-      basePath: base,
-      milestoneId: "M001",
-      ctx: mkCtx(cap),
-      pi: mkPiThrowOnce(cap),
-    });
-
-    checkAutoStartAfterDiscuss(); // sendMessage throws → count stays 0
-
-    const entryAfterThrow = _getPendingAutoStart(base);
-    assert.equal(entryAfterThrow!.planBlockedRecoveryCount, 0, "count is 0 after throw");
-
-    checkAutoStartAfterDiscuss(); // sendMessage succeeds → count becomes 1
-    assert.equal(cap.messages.length, 1, "second call must produce one successful sendMessage");
-
-    const entryAfterSuccess = _getPendingAutoStart(base);
-    assert.equal(
-      entryAfterSuccess!.planBlockedRecoveryCount,
-      1,
-      "counter must be 1 after first successful dispatch",
-    );
-  });
-
-  test("3 successful sendMessage calls exhaust the budget; 4th emits escalation notify", () => {
-    base = mkBase();
-    openDatabase(":memory:");
-    insertMilestone({ id: "M001", title: "Corrections Test", status: "queued" });
-
-    cap = mkCapture();
-    setPendingAutoStart(base, {
-      basePath: base,
-      milestoneId: "M001",
-      ctx: mkCtx(cap),
-      pi: mkPi(cap),
-    });
-
-    // Three successful recoveries
-    checkAutoStartAfterDiscuss(); // count → 1
-    checkAutoStartAfterDiscuss(); // count → 2
-    checkAutoStartAfterDiscuss(); // count → 3
-
-    const entry = _getPendingAutoStart(base);
-    assert.equal(entry!.planBlockedRecoveryCount, 3, "counter must be 3 after three successes");
-    assert.equal(cap.messages.length, 3, "three sendMessage calls must have occurred");
-
-    // Fourth call hits the cap
-    cap.notifies = [];
-    cap.messages = [];
-    const resultAtCap = checkAutoStartAfterDiscuss();
-    assert.equal(resultAtCap, false, "4th call must return false");
-    assert.equal(cap.messages.length, 0, "4th call must NOT call sendMessage");
-    const errorNotify = cap.notifies.find((n) => n.level === "error");
-    assert.ok(errorNotify, "4th call must emit escalation notify with level 'error'");
-  });
-});

package/src/resources/extensions/gsd/tests/gate-1b-recovery-bound.test.ts

@@ -1,218 +0,0 @@
-// gsd-pi + Gate 1b recovery counter bound — regression tests for H1 fix (#5012)
-//
-// Verifies that checkAutoStartAfterDiscuss stops emitting plan-blocked recovery
-// hints (with triggerTurn:true) after MAX_PLAN_BLOCKED_RECOVERIES attempts and
-// instead escalates to the user via ctx.ui.notify("error"), breaking the loop.
-
-import { describe, test, beforeEach, afterEach } from "node:test";
-import assert from "node:assert/strict";
-import { mkdtempSync, mkdirSync, rmSync, writeFileSync } from "node:fs";
-import { join } from "node:path";
-import { tmpdir } from "node:os";
-
-import {
-  checkAutoStartAfterDiscuss,
-  setPendingAutoStart,
-  clearPendingAutoStart,
-  _getPendingAutoStart,
-} from "../guided-flow.ts";
-import { drainLogs } from "../workflow-logger.ts";
-import {
-  openDatabase,
-  closeDatabase,
-  insertMilestone,
-} from "../gsd-db.ts";
-
-// ─── Harness ───────────────────────────────────────────────────────────────
-
-interface MockCapture {
-  notifies: Array<{ msg: string; level: string }>;
-  messages: Array<{ payload: any; options: any }>;
-}
-
-function mkCapture(): MockCapture {
-  return { notifies: [], messages: [] };
-}
-
-function mkCtx(cap: MockCapture): any {
-  return {
-    ui: {
-      notify: (msg: string, level: string) => {
-        cap.notifies.push({ msg, level });
-      },
-    },
-  };
-}
-
-function mkPi(cap: MockCapture): any {
-  return {
-    sendMessage: (payload: any, options: any) => {
-      cap.messages.push({ payload, options });
-    },
-    setActiveTools: () => undefined,
-    getActiveTools: () => [],
-  };
-}
-
-function mkBase(): string {
-  const base = mkdtempSync(join(tmpdir(), "gsd-gate1b-bound-"));
-  mkdirSync(join(base, ".gsd", "milestones", "M001"), { recursive: true });
-  writeFileSync(
-    join(base, ".gsd", "milestones", "M001", "M001-CONTEXT.md"),
-    "# M001: Bound Test\n\nContext written by discuss phase.\n",
-  );
-  return base;
-}
-
-// ─── Tests ─────────────────────────────────────────────────────────────────
-
-describe("Gate 1b recovery bound (H1)", () => {
-  let base: string;
-  let cap: MockCapture;
-
-  beforeEach(() => {
-    clearPendingAutoStart();
-    drainLogs();
-  });
-
-  afterEach(() => {
-    closeDatabase();
-    clearPendingAutoStart();
-    if (base) {
-      rmSync(base, { recursive: true, force: true });
-    }
-  });
-
-  test("first N-1 invocations increment counter and emit recovery with triggerTurn:true", () => {
-    base = mkBase();
-    openDatabase(":memory:");
-    insertMilestone({ id: "M001", title: "Bound Test", status: "queued" });
-
-    cap = mkCapture();
-    setPendingAutoStart(base, {
-      basePath: base,
-      milestoneId: "M001",
-      ctx: mkCtx(cap),
-      pi: mkPi(cap),
-    });
-
-    // MAX_PLAN_BLOCKED_RECOVERIES = 3; first two calls should emit recovery
-    const resultOne = checkAutoStartAfterDiscuss();
-    assert.equal(resultOne, false, "call 1: must return false");
-    assert.equal(cap.messages.length, 1, "call 1: exactly one sendMessage");
-    assert.equal(cap.messages[0].options.triggerTurn, true, "call 1: triggerTurn must be true");
-    assert.equal(cap.messages[0].payload.customType, "gsd-plan-milestone-blocked-recovery");
-
-    const entryAfterOne = _getPendingAutoStart(base);
-    assert.ok(entryAfterOne, "entry must still exist after call 1");
-    assert.equal(entryAfterOne.planBlockedRecoveryCount, 1, "counter must be 1 after call 1");
-
-    const resultTwo = checkAutoStartAfterDiscuss();
-    assert.equal(resultTwo, false, "call 2: must return false");
-    assert.equal(cap.messages.length, 2, "call 2: second sendMessage emitted");
-    assert.equal(cap.messages[1].options.triggerTurn, true, "call 2: triggerTurn must be true");
-
-    const entryAfterTwo = _getPendingAutoStart(base);
-    assert.ok(entryAfterTwo, "entry must still exist after call 2");
-    assert.equal(entryAfterTwo.planBlockedRecoveryCount, 2, "counter must be 2 after call 2");
-  });
-
-  test("Nth invocation (at MAX_PLAN_BLOCKED_RECOVERIES) escalates via notify(error) without sendMessage(triggerTurn)", () => {
-    base = mkBase();
-    openDatabase(":memory:");
-    insertMilestone({ id: "M001", title: "Bound Test", status: "queued" });
-
-    cap = mkCapture();
-    setPendingAutoStart(base, {
-      basePath: base,
-      milestoneId: "M001",
-      ctx: mkCtx(cap),
-      pi: mkPi(cap),
-    });
-
-    // Exhaust the recovery budget (MAX = 3): call 3 times to reach the limit
-    checkAutoStartAfterDiscuss(); // count → 1
-    checkAutoStartAfterDiscuss(); // count → 2
-    checkAutoStartAfterDiscuss(); // count → 3
-
-    // At count = 3 the counter equals MAX so the next call must escalate
-    cap.messages = [];
-    cap.notifies = [];
-    drainLogs();
-
-    const resultAtLimit = checkAutoStartAfterDiscuss();
-    assert.equal(resultAtLimit, false, "at-limit call: must return false");
-
-    // Must NOT trigger a new LLM turn
-    assert.equal(
-      cap.messages.length,
-      0,
-      "at-limit call: sendMessage must NOT be called (loop must stop)",
-    );
-    const triggerMessages = cap.messages.filter((m) => m.options?.triggerTurn);
-    assert.equal(triggerMessages.length, 0, "no triggerTurn message after limit");
-
-    // Must escalate to user via notify("error")
-    const errorNotify = cap.notifies.find((n) => n.level === "error");
-    assert.ok(errorNotify, "at-limit call: ctx.ui.notify('error') must be called");
-    assert.match(
-      errorNotify.msg,
-      /gsd-debug/i,
-      "error notification must direct user to run /gsd-debug",
-    );
-    assert.match(
-      errorNotify.msg,
-      /M001/,
-      "error notification must include the milestone ID",
-    );
-
-    // Confirm the log records the escalation
-    const logs = drainLogs();
-    const escalationLog = logs.find(
-      (e) => e.component === "guided" && /Gate 1b/.test(e.message) && /escalat/.test(e.message),
-    );
-    assert.ok(escalationLog, "escalation must be logged via logWarning");
-  });
-
-  test("after clearPendingAutoStart + setPendingAutoStart the counter is reset to 0", () => {
-    base = mkBase();
-    openDatabase(":memory:");
-    insertMilestone({ id: "M001", title: "Bound Test", status: "queued" });
-
-    cap = mkCapture();
-    setPendingAutoStart(base, {
-      basePath: base,
-      milestoneId: "M001",
-      ctx: mkCtx(cap),
-      pi: mkPi(cap),
-    });
-
-    // Advance counter to 2
-    checkAutoStartAfterDiscuss();
-    checkAutoStartAfterDiscuss();
-
-    const entryBefore = _getPendingAutoStart(base);
-    assert.ok(entryBefore, "entry must exist");
-    assert.equal(entryBefore.planBlockedRecoveryCount, 2, "counter must be 2 before reset");
-
-    // Simulate user retry: clear then re-set
-    clearPendingAutoStart(base);
-    cap = mkCapture();
-    setPendingAutoStart(base, {
-      basePath: base,
-      milestoneId: "M001",
-      ctx: mkCtx(cap),
-      pi: mkPi(cap),
-    });
-
-    const entryAfter = _getPendingAutoStart(base);
-    assert.ok(entryAfter, "entry must exist after re-set");
-    assert.equal(entryAfter.planBlockedRecoveryCount, 0, "counter must be 0 after re-set (fresh entry)");
-
-    // Verify first call after reset emits recovery, not escalation
-    const result = checkAutoStartAfterDiscuss();
-    assert.equal(result, false, "first call after reset must return false");
-    assert.equal(cap.messages.length, 1, "recovery hint must be emitted after reset");
-    assert.equal(cap.messages[0].options.triggerTurn, true, "triggerTurn must be true after reset");
-  });
-});