gsd-pi 2.76.0 → 2.77.0-dev.1d17f366c

package/src/resources/extensions/gsd/tests/ready-phrase-no-files-4573.test.ts

@@ -0,0 +1,388 @@
+/**
+ * GSD-2 / guided-flow — regression tests for #4573
+ *
+ * Covers two recovery paths:
+ *   - maybeHandleReadyPhraseWithoutFiles: nudge when LLM emits
+ *     "Milestone M001 ready." without writing CONTEXT.md / ROADMAP.md
+ *   - maybeHandleEmptyIntentTurn: nudge when LLM narrates intent but
+ *     emits no tool-use blocks
+ */
+
+import { describe, test, beforeEach } 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 {
+  setPendingAutoStart,
+  clearPendingAutoStart,
+  maybeHandleReadyPhraseWithoutFiles,
+  maybeHandleEmptyIntentTurn,
+  resetEmptyTurnCounter,
+} from "../guided-flow.ts";
+
+// ─── Test 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, opts: { sendThrows?: boolean } = {}): any {
+  return {
+    sendMessage: (payload: any, options: any) => {
+      if (opts.sendThrows) throw new Error("send failed");
+      cap.messages.push({ payload, options });
+    },
+    setActiveTools: () => undefined,
+    getActiveTools: () => [],
+  };
+}
+
+function mkBase(): string {
+  const base = mkdtempSync(join(tmpdir(), "gsd-4573-"));
+  mkdirSync(join(base, ".gsd", "milestones", "M001"), { recursive: true });
+  return base;
+}
+
+function assistantMsg(text: string, opts: { toolUse?: boolean } = {}): any {
+  const content: any[] = [];
+  if (text) content.push({ type: "text", text });
+  if (opts.toolUse) content.push({ type: "tool_use", name: "whatever", input: {} });
+  return { role: "assistant", content };
+}
+
+// ─── ready-phrase recovery (Layer 2) ───────────────────────────────────────
+
+describe("#4573 maybeHandleReadyPhraseWithoutFiles", () => {
+  beforeEach(() => {
+    clearPendingAutoStart();
+    resetEmptyTurnCounter();
+  });
+
+  test("no pending entry → no-op", () => {
+    const cap = mkCapture();
+    const event = { messages: [assistantMsg("Milestone M001 ready.")] };
+    const handled = maybeHandleReadyPhraseWithoutFiles(event);
+    assert.equal(handled, false);
+    assert.equal(cap.messages.length, 0);
+  });
+
+  test("pending entry, ready phrase, no files → notify + sendMessage", () => {
+    const base = mkBase();
+    try {
+      const cap = mkCapture();
+      setPendingAutoStart(base, {
+        basePath: base,
+        milestoneId: "M001",
+        ctx: mkCtx(cap),
+        pi: mkPi(cap),
+      });
+      const handled = maybeHandleReadyPhraseWithoutFiles({
+        messages: [assistantMsg("Milestone M001 ready.")],
+      });
+      assert.equal(handled, true);
+      assert.equal(cap.messages.length, 1);
+      assert.equal(cap.messages[0].payload.customType, "gsd-ready-no-files");
+      assert.equal(cap.messages[0].options.triggerTurn, true);
+      assert.ok(
+        cap.notifies.some((n) => /rejected/.test(n.msg)),
+        "user notified about rejection",
+      );
+    } finally {
+      clearPendingAutoStart();
+    }
+  });
+
+  test("retry cap — after MAX_READY_REJECTS the nudge stops and entry clears", () => {
+    const base = mkBase();
+    try {
+      const cap = mkCapture();
+      setPendingAutoStart(base, {
+        basePath: base,
+        milestoneId: "M001",
+        ctx: mkCtx(cap),
+        pi: mkPi(cap),
+      });
+      const event = { messages: [assistantMsg("Milestone M001 ready.")] };
+
+      const first = maybeHandleReadyPhraseWithoutFiles(event);
+      const second = maybeHandleReadyPhraseWithoutFiles(event);
+      const third = maybeHandleReadyPhraseWithoutFiles(event); // > MAX
+
+      assert.equal(first, true);
+      assert.equal(second, true);
+      assert.equal(third, true); // still returns true (handled via give-up)
+      assert.equal(cap.messages.length, 2, "only 2 nudges sent (MAX_READY_REJECTS=2)");
+      assert.ok(
+        cap.notifies.some((n) => /Stopping auto-nudge/.test(n.msg)),
+        "gives up with error notify",
+      );
+
+      // After giving up, a fresh re-entry starts clean
+      const fourth = maybeHandleReadyPhraseWithoutFiles(event);
+      assert.equal(fourth, false, "pending entry was cleared — nothing to handle");
+    } finally {
+      clearPendingAutoStart();
+    }
+  });
+
+  test("files present → no nudge (happy path already fired)", () => {
+    const base = mkBase();
+    try {
+      writeFileSync(join(base, ".gsd", "milestones", "M001", "M001-CONTEXT.md"), "# ctx");
+      const cap = mkCapture();
+      setPendingAutoStart(base, {
+        basePath: base,
+        milestoneId: "M001",
+        ctx: mkCtx(cap),
+        pi: mkPi(cap),
+      });
+      const handled = maybeHandleReadyPhraseWithoutFiles({
+        messages: [assistantMsg("Milestone M001 ready.")],
+      });
+      assert.equal(handled, false);
+      assert.equal(cap.messages.length, 0);
+    } finally {
+      clearPendingAutoStart();
+    }
+  });
+
+  test("last message lacks ready phrase → no-op", () => {
+    const base = mkBase();
+    try {
+      const cap = mkCapture();
+      setPendingAutoStart(base, {
+        basePath: base,
+        milestoneId: "M001",
+        ctx: mkCtx(cap),
+        pi: mkPi(cap),
+      });
+      const handled = maybeHandleReadyPhraseWithoutFiles({
+        messages: [assistantMsg("Let me think about the slices first.")],
+      });
+      assert.equal(handled, false);
+      assert.equal(cap.messages.length, 0);
+    } finally {
+      clearPendingAutoStart();
+    }
+  });
+
+  test("fresh entry after give-up resets counter", () => {
+    const base = mkBase();
+    try {
+      const cap = mkCapture();
+      // First cycle: exhaust cap
+      setPendingAutoStart(base, {
+        basePath: base,
+        milestoneId: "M001",
+        ctx: mkCtx(cap),
+        pi: mkPi(cap),
+      });
+      const event = { messages: [assistantMsg("Milestone M001 ready.")] };
+      maybeHandleReadyPhraseWithoutFiles(event);
+      maybeHandleReadyPhraseWithoutFiles(event);
+      maybeHandleReadyPhraseWithoutFiles(event); // clears entry
+
+      // New /gsd run — re-seeds entry; counter must be 0 again
+      cap.messages.length = 0;
+      setPendingAutoStart(base, {
+        basePath: base,
+        milestoneId: "M001",
+        ctx: mkCtx(cap),
+        pi: mkPi(cap),
+      });
+      const handled = maybeHandleReadyPhraseWithoutFiles(event);
+      assert.equal(handled, true);
+      assert.equal(cap.messages.length, 1, "fresh entry fires nudge again");
+    } finally {
+      clearPendingAutoStart();
+    }
+  });
+});
+
+// ─── empty-turn recovery (Layer 3) ────────────────────────────────────────
+
+describe("#4573 maybeHandleEmptyIntentTurn", () => {
+  beforeEach(() => {
+    clearPendingAutoStart();
+    resetEmptyTurnCounter();
+  });
+
+  test("no pending entry + isAuto false → no-op (interactive discuss is user-driven)", () => {
+    const event = { messages: [assistantMsg("I'll write the CONTEXT.md now.")] };
+    const handled = maybeHandleEmptyIntentTurn(event, false);
+    assert.equal(handled, false);
+  });
+
+  test("text-only turn WITHOUT commit phrase → not flagged (legitimate text)", () => {
+    const base = mkBase();
+    try {
+      const cap = mkCapture();
+      setPendingAutoStart(base, {
+        basePath: base,
+        milestoneId: "M001",
+        ctx: mkCtx(cap),
+        pi: mkPi(cap),
+      });
+      const handled = maybeHandleEmptyIntentTurn(
+        { messages: [assistantMsg("Here is the roadmap preview — three slices.")] },
+        false,
+      );
+      assert.equal(handled, false);
+      assert.equal(cap.messages.length, 0);
+    } finally {
+      clearPendingAutoStart();
+    }
+  });
+
+  test("text-only turn ending in question → treated as user-handoff, not flagged", () => {
+    const base = mkBase();
+    try {
+      const cap = mkCapture();
+      setPendingAutoStart(base, {
+        basePath: base,
+        milestoneId: "M001",
+        ctx: mkCtx(cap),
+        pi: mkPi(cap),
+      });
+      const handled = maybeHandleEmptyIntentTurn(
+        { messages: [assistantMsg("Ready to write, or want to adjust?")] },
+        false,
+      );
+      assert.equal(handled, false);
+    } finally {
+      clearPendingAutoStart();
+    }
+  });
+
+  test("commit-intent phrase WITHOUT tool call → nudge fires", () => {
+    const base = mkBase();
+    try {
+      const cap = mkCapture();
+      setPendingAutoStart(base, {
+        basePath: base,
+        milestoneId: "M001",
+        ctx: mkCtx(cap),
+        pi: mkPi(cap),
+      });
+      const handled = maybeHandleEmptyIntentTurn(
+        { messages: [assistantMsg("I'll now write the CONTEXT.md file.")] },
+        false,
+      );
+      assert.equal(handled, true);
+      assert.equal(cap.messages.length, 1);
+      assert.equal(cap.messages[0].payload.customType, "gsd-empty-turn-recovery");
+    } finally {
+      clearPendingAutoStart();
+    }
+  });
+
+  test("commit-intent WITH tool-use block → not flagged", () => {
+    const base = mkBase();
+    try {
+      const cap = mkCapture();
+      setPendingAutoStart(base, {
+        basePath: base,
+        milestoneId: "M001",
+        ctx: mkCtx(cap),
+        pi: mkPi(cap),
+      });
+      const handled = maybeHandleEmptyIntentTurn(
+        { messages: [assistantMsg("I'll write the file now.", { toolUse: true })] },
+        false,
+      );
+      assert.equal(handled, false);
+      assert.equal(cap.messages.length, 0);
+    } finally {
+      clearPendingAutoStart();
+    }
+  });
+
+  test("ready phrase is NOT treated as empty-turn (handled by other recovery path)", () => {
+    const base = mkBase();
+    try {
+      const cap = mkCapture();
+      setPendingAutoStart(base, {
+        basePath: base,
+        milestoneId: "M001",
+        ctx: mkCtx(cap),
+        pi: mkPi(cap),
+      });
+      const handled = maybeHandleEmptyIntentTurn(
+        { messages: [assistantMsg("Milestone M001 ready.")] },
+        false,
+      );
+      assert.equal(handled, false);
+    } finally {
+      clearPendingAutoStart();
+    }
+  });
+
+  test("empty-turn retry cap — stops after MAX_EMPTY_TURN_RETRIES", () => {
+    const base = mkBase();
+    try {
+      const cap = mkCapture();
+      setPendingAutoStart(base, {
+        basePath: base,
+        milestoneId: "M001",
+        ctx: mkCtx(cap),
+        pi: mkPi(cap),
+      });
+      const event = { messages: [assistantMsg("I'll write the CONTEXT.md file.")] };
+
+      maybeHandleEmptyIntentTurn(event, false); // 1
+      maybeHandleEmptyIntentTurn(event, false); // 2
+      const third = maybeHandleEmptyIntentTurn(event, false); // > cap
+
+      assert.equal(cap.messages.length, 2, "only 2 nudges sent");
+      assert.equal(third, false, "after cap, no further injection");
+      assert.ok(
+        cap.notifies.some((n) => /Stopping auto-nudge/.test(n.msg)),
+        "user notified of give-up",
+      );
+    } finally {
+      clearPendingAutoStart();
+    }
+  });
+
+  test("resetEmptyTurnCounter clears state after a successful tool-use turn", () => {
+    const base = mkBase();
+    try {
+      const cap = mkCapture();
+      setPendingAutoStart(base, {
+        basePath: base,
+        milestoneId: "M001",
+        ctx: mkCtx(cap),
+        pi: mkPi(cap),
+      });
+      const event = { messages: [assistantMsg("I'll write the CONTEXT.md file.")] };
+
+      maybeHandleEmptyIntentTurn(event, false); // 1
+      maybeHandleEmptyIntentTurn(event, false); // 2 — at cap
+      resetEmptyTurnCounter(); // simulate a successful tool-use turn in between
+
+      cap.messages.length = 0;
+      const after = maybeHandleEmptyIntentTurn(event, false);
+      assert.equal(after, true, "counter reset — nudge fires again");
+      assert.equal(cap.messages.length, 1);
+    } finally {
+      clearPendingAutoStart();
+    }
+  });
+});

package/src/resources/extensions/gsd/tests/require-slice-discussion-dispatch.test.ts

@@ -0,0 +1,170 @@
+/**
+ * Regression tests for #3454: auto-dispatch must honour
+ * require_slice_discussion and pause before plan-slice when:
+ *   1. state.phase === "planning"
+ *   2. require_slice_discussion is enabled in preferences
+ *   3. state.activeSlice is non-null
+ *   4. the slice has no CONTEXT file on disk
+ *
+ * Exercises the dispatch rule from DISPATCH_RULES directly with a
+ * DispatchContext built against a real temp directory — no source-string
+ * matching, no brittle rename dependencies.
+ */
+import { test, describe } from "node:test";
+import assert from "node:assert/strict";
+import { mkdtempSync, mkdirSync, writeFileSync, rmSync } from "node:fs";
+import { join } from "node:path";
+import { tmpdir } from "node:os";
+import { DISPATCH_RULES, type DispatchContext } from "../auto-dispatch.ts";
+import type { GSDState } from "../types.ts";
+import type { GSDPreferences } from "../preferences.ts";
+
+// Use a stable token (the preference name) for rule lookup instead of the
+// full human-readable title — copy edits to the rule name will not break
+// these tests as long as the rule still references the preference it gates.
+const RULE_NAME_TOKEN = "require_slice_discussion";
+
+function findRule() {
+  const matches = DISPATCH_RULES.filter(r => r.name.includes(RULE_NAME_TOKEN));
+  if (matches.length !== 1) {
+    throw new Error(`expected exactly one dispatch rule containing "${RULE_NAME_TOKEN}", found ${matches.length}`);
+  }
+  return matches[0];
+}
+
+function buildState(overrides: Partial<GSDState> = {}): GSDState {
+  return {
+    activeMilestone: { id: "M001", title: "Test milestone" },
+    activeSlice: { id: "S01", title: "Test slice" },
+    activeTask: null,
+    phase: "planning",
+    recentDecisions: [],
+    blockers: [],
+    nextAction: "",
+    registry: [],
+    ...overrides,
+  };
+}
+
+function makeBasePath(prefix: string): string {
+  const dir = mkdtempSync(join(tmpdir(), `gsd-req-slice-${prefix}-`));
+  mkdirSync(join(dir, ".gsd", "milestones", "M001", "slices", "S01"), { recursive: true });
+  return dir;
+}
+
+function buildCtx(
+  basePath: string,
+  prefs: GSDPreferences | undefined,
+  state: GSDState = buildState(),
+): DispatchContext {
+  return {
+    basePath,
+    mid: "M001",
+    midTitle: "Test milestone",
+    state,
+    prefs,
+  };
+}
+
+describe("require_slice_discussion dispatch rule (#3454)", () => {
+  // ─── Positive case: rule fires ────────────────────────────────────────
+
+  test("returns stop action when preference enabled and CONTEXT missing", async () => {
+    const basePath = makeBasePath("fire");
+    try {
+      const prefs = { phases: { require_slice_discussion: true } } as unknown as GSDPreferences;
+      const action = await findRule().match(buildCtx(basePath, prefs));
+      assert.ok(action, "rule must return a non-null action when pausing is required");
+      assert.strictEqual(action!.action, "stop");
+      if (action!.action === "stop") {
+        assert.match(action!.reason, /S01/);
+        assert.match(action!.reason, /require_slice_discussion/);
+        assert.match(action!.reason, /\/gsd discuss/);
+        assert.strictEqual(action!.level, "info");
+      }
+    } finally {
+      rmSync(basePath, { recursive: true, force: true });
+    }
+  });
+
+  // ─── Negative cases: rule falls through ───────────────────────────────
+
+  test("falls through (null) when preference is disabled", async () => {
+    const basePath = makeBasePath("pref-disabled");
+    try {
+      const prefs = { phases: { require_slice_discussion: false } } as unknown as GSDPreferences;
+      const action = await findRule().match(buildCtx(basePath, prefs));
+      assert.strictEqual(action, null);
+    } finally {
+      rmSync(basePath, { recursive: true, force: true });
+    }
+  });
+
+  test("falls through (null) when preference is absent / undefined", async () => {
+    const basePath = makeBasePath("pref-absent");
+    try {
+      const action = await findRule().match(buildCtx(basePath, undefined));
+      assert.strictEqual(action, null);
+    } finally {
+      rmSync(basePath, { recursive: true, force: true });
+    }
+  });
+
+  test("falls through (null) when phase is not 'planning'", async () => {
+    const basePath = makeBasePath("wrong-phase");
+    try {
+      const prefs = { phases: { require_slice_discussion: true } } as unknown as GSDPreferences;
+      const state = buildState({ phase: "executing" });
+      const action = await findRule().match(buildCtx(basePath, prefs, state));
+      assert.strictEqual(action, null);
+    } finally {
+      rmSync(basePath, { recursive: true, force: true });
+    }
+  });
+
+  test("falls through (null) when no active slice", async () => {
+    const basePath = makeBasePath("no-slice");
+    try {
+      const prefs = { phases: { require_slice_discussion: true } } as unknown as GSDPreferences;
+      const state = buildState({ activeSlice: null });
+      const action = await findRule().match(buildCtx(basePath, prefs, state));
+      assert.strictEqual(action, null);
+    } finally {
+      rmSync(basePath, { recursive: true, force: true });
+    }
+  });
+
+  // ─── Context-file present: should not pause ───────────────────────────
+
+  test("falls through (null) when CONTEXT file already exists on disk", async () => {
+    const basePath = makeBasePath("ctx-present");
+    try {
+      // Seed the CONTEXT file that /gsd discuss would have written.
+      const sliceDir = join(basePath, ".gsd", "milestones", "M001", "slices", "S01");
+      writeFileSync(join(sliceDir, "S01-CONTEXT.md"), "# Discussion notes\n", "utf-8");
+
+      const prefs = { phases: { require_slice_discussion: true } } as unknown as GSDPreferences;
+      const action = await findRule().match(buildCtx(basePath, prefs));
+      assert.strictEqual(
+        action,
+        null,
+        "once the slice has a CONTEXT file, the rule must fall through so planning proceeds",
+      );
+    } finally {
+      rmSync(basePath, { recursive: true, force: true });
+    }
+  });
+
+  // ─── Rule ordering: must run before the plan-slice rule ──────────────
+
+  test("rule is ordered before the 'planning → plan-slice' rule", () => {
+    const discussIdx = DISPATCH_RULES.findIndex(r => r.name.includes(RULE_NAME_TOKEN));
+    const planIdx = DISPATCH_RULES.findIndex(r => r.name.startsWith("planning → plan-slice"));
+    assert.ok(discussIdx >= 0, "require_slice_discussion rule must be registered");
+    assert.ok(planIdx >= 0, "plan-slice rule must be registered");
+    assert.ok(
+      discussIdx < planIdx,
+      "require_slice_discussion rule must be ordered before plan-slice so it preempts dispatch",
+    );
+  });
+});

package/src/resources/extensions/gsd/tests/restore-tools-after-discuss.test.ts

@@ -45,9 +45,15 @@ describe('restore tools after discuss flow scoping (#3628)', () => {
   })
 
   it('savedTools is restored after sendMessage', () => {
-    // Find the sendMessage call
-    const sendMsg = src.indexOf('triggerTurn: true')
-    assert.ok(sendMsg !== -1, 'sendMessage with triggerTurn must exist')
+    // #4573: guided-flow.ts now contains multiple `triggerTurn: true` calls
+    // (ready-phrase and empty-turn recovery paths). The discuss-flow scoping
+    // sendMessage is the one that follows `savedTools = currentTools`, so
+    // anchor the search there rather than at the first `triggerTurn: true`.
+    const savedToolsAssign = src.indexOf('savedTools = currentTools')
+    assert.ok(savedToolsAssign !== -1, 'savedTools = currentTools must exist')
+
+    const sendMsg = src.indexOf('triggerTurn: true', savedToolsAssign)
+    assert.ok(sendMsg !== -1, 'discuss-flow sendMessage with triggerTurn must exist after savedTools capture')
 
     // After sendMessage, savedTools should be restored via setActiveTools
     const afterSend = src.slice(sendMsg, sendMsg + 500)