gsd-pi 2.77.0 → 2.78.0

package/src/resources/extensions/gsd/tests/context-store.test.ts

@@ -627,4 +627,83 @@ Integration tests mock external services.
 
     assert.strictEqual(result, '', 'empty content returns empty string');
   });
+
+  // ── Regression: issue #4719 — single-H2 with many H3 entries ──────────────
+  // A KNOWLEDGE.md structured as one top-level H2 with many H3 entries must
+  // filter at H3 granularity; otherwise one keyword match against the H2
+  // header or first paragraph returns the entire file.
+  test("single H2 with many H3 entries filters at H3 level (issue #4719)", async () => {
+    const singleH2Knowledge = `# Project Knowledge
+
+## Patterns
+
+### Database: prepared statements
+Always use prepared statements with SQLite.
+
+### API: versioned paths
+Use /v1/resource style versioning.
+
+### Testing: node:test
+Prefer node:test over external frameworks.
+
+### Deployment: blue-green
+Blue-green deployment for zero-downtime releases.
+`;
+
+    const result = await queryKnowledge(singleH2Knowledge, ['database']);
+
+    // Should include only the matching H3 entry, not the whole file
+    assert.match(result, /Database: prepared statements/, 'includes matching H3 entry');
+    assert.ok(
+      !result.includes('API: versioned paths'),
+      'does not include non-matching H3 entry',
+    );
+    assert.ok(
+      !result.includes('Testing: node:test'),
+      'does not include non-matching H3 entry',
+    );
+    assert.ok(
+      !result.includes('Deployment: blue-green'),
+      'does not include non-matching H3 entry',
+    );
+    // The returned payload must be dramatically smaller than the full content
+    assert.ok(
+      result.length < singleH2Knowledge.length / 2,
+      `scoped result (${result.length} chars) should be <50% of full content (${singleH2Knowledge.length} chars)`,
+    );
+  });
+
+  test("single H2 with H3 entries returns empty when no H3 matches (issue #4719)", async () => {
+    const singleH2Knowledge = `# Project Knowledge
+
+## Patterns
+
+### Database: prepared statements
+Always use prepared statements with SQLite.
+
+### API: versioned paths
+Use /v1/resource style versioning.
+`;
+
+    const result = await queryKnowledge(singleH2Knowledge, ['nonexistent']);
+
+    assert.strictEqual(result, '', 'no H3 match returns empty string');
+  });
+
+  test("falls back to H2 when no H3 headings exist at all", async () => {
+    // Backwards-compat: files with only H2 topic headers must still filter.
+    const h2OnlyKnowledge = `# Project Knowledge
+
+## Database Patterns
+Use prepared statements.
+
+## API Design
+REST with OpenAPI.
+`;
+
+    const result = await queryKnowledge(h2OnlyKnowledge, ['database']);
+
+    assert.match(result, /Database Patterns/, 'H2-only file falls back to H2 filtering');
+    assert.ok(!result.includes('API Design'), 'non-matching H2 section excluded');
+  });
 });

package/src/resources/extensions/gsd/tests/copy-planning-artifacts-samepath.test.ts

@@ -2,6 +2,7 @@ import test from "node:test";
 import assert from "node:assert/strict";
 import { readFileSync } from "node:fs";
 import { join } from "node:path";
+import { extractSourceRegion } from "./test-helpers.ts";
 
 test("copyPlanningArtifacts skips when source and destination .gsd resolve to the same path", () => {
   const srcPath = join(import.meta.dirname, "..", "auto-worktree.ts");
@@ -10,7 +11,7 @@ test("copyPlanningArtifacts skips when source and destination .gsd resolve to th
   const fnIdx = src.indexOf("function copyPlanningArtifacts");
   assert.ok(fnIdx !== -1, "copyPlanningArtifacts function exists");
 
-  const fnBody = src.slice(fnIdx, fnIdx + 2400);
+  const fnBody = extractSourceRegion(src, "function copyPlanningArtifacts");
 
   const guardIdx = fnBody.indexOf("if (isSamePath(srcGsd, dstGsd)) return;");
   const copyIdx = fnBody.indexOf("safeCopyRecursive(join(srcGsd, \"milestones\")");

package/src/resources/extensions/gsd/tests/crash-recovery.test.ts

@@ -1,6 +1,6 @@
 import test from "node:test";
 import assert from "node:assert/strict";
-import { mkdirSync, rmSync, writeFileSync } from "node:fs";
+import { existsSync, mkdirSync, rmSync, writeFileSync } from "node:fs";
 import { join } from "node:path";
 import { tmpdir } from "node:os";
 import { randomUUID } from "node:crypto";
@@ -21,6 +21,7 @@ import {
 } from "../interrupted-session.ts";
 import { gsdRoot } from "../paths.ts";
 import type { GSDState } from "../types.ts";
+import { _synthesizePausedSessionRecoveryForTest } from "../auto.ts";
 
 function makeTmpBase(): string {
   const base = join(tmpdir(), `gsd-test-${randomUUID()}`);
@@ -167,6 +168,54 @@ test("readPausedSessionMetadata reads paused-session metadata when present", ()
   }
 });
 
+test("paused session recovery consumes JSONL without deleting the evidence file", (t) => {
+  const base = makeTmpBase();
+  t.after(() => cleanup(base));
+
+  const sessionFile = join(base, "paused-session.jsonl");
+  writeFileSync(
+    sessionFile,
+    [
+      JSON.stringify({ type: "session", id: "session-1" }),
+      JSON.stringify({
+        type: "message",
+        message: {
+          role: "assistant",
+          content: [
+            {
+              type: "toolCall",
+              name: "bash",
+              id: "tool-1",
+              arguments: { command: "echo paused" },
+            },
+          ],
+        },
+      }),
+      JSON.stringify({
+        type: "message",
+        message: {
+          role: "toolResult",
+          toolCallId: "tool-1",
+          toolName: "bash",
+          isError: false,
+          content: "paused\n",
+        },
+      }),
+    ].join("\n"),
+    "utf-8",
+  );
+
+  const recovery = _synthesizePausedSessionRecoveryForTest(
+    base,
+    "execute-task",
+    "M001/S01/T01",
+    sessionFile,
+  );
+
+  assert.equal(recovery?.trace.toolCallCount, 1);
+  assert.equal(existsSync(sessionFile), true, "paused JSONL must remain available after synthesis");
+});
+
 test("readPausedSessionMetadata preserves unitType and unitId through round-trip", () => {
   const base = makeTmpBase();
   try {

package/src/resources/extensions/gsd/tests/custom-engine-loop-integration.test.ts

@@ -481,6 +481,165 @@ describe("Custom engine loop integration", () => {
     );
   });
 
+  it("stops custom workflow after repeated verification retries", async () => {
+    _resetPendingResolve();
+
+    const runDir = makeTmpDir();
+    const graph = makeGraph([makeStep({ id: "retry-step" })], "retry-exhaustion");
+    writeGraph(runDir, graph);
+    writeFileSync(join(runDir, "DEFINITION.yaml"), stringify({
+      version: 1,
+      name: "retry-exhaustion",
+      steps: [{
+        id: "retry-step",
+        name: "retry-step",
+        prompt: "Do retry-step",
+        produces: "retry-step/output.md",
+        verify: { policy: "shell-command", command: "exit 1" },
+      }],
+    }));
+
+    const ctx = makeMockCtx();
+    const pi = makeMockPi();
+    const s = makeLoopSession({
+      activeEngineId: "custom",
+      activeRunDir: runDir,
+      basePath: runDir,
+    });
+    const deps = makeMockDeps({
+      stopAuto: async (_ctx, _pi, reason) => {
+        deps.callLog.push(`stopAuto:${reason ?? "no-reason"}`);
+        s.active = false;
+      },
+    });
+
+    const resolver = setInterval(() => {
+      resolveAgentEnd({ messages: [{ role: "assistant" }] });
+    }, 25);
+    let timeout: NodeJS.Timeout | undefined;
+    try {
+      await Promise.race([
+        autoLoop(ctx, pi, s, deps),
+        new Promise((_, reject) =>
+          timeout = setTimeout(() => {
+            s.active = false;
+            resolveAgentEnd({ messages: [{ role: "assistant" }] });
+            reject(new Error(
+              `autoLoop did not stop after verification retry exhaustion; calls=${pi.calls.length}; log=${deps.callLog.join(",")}`,
+            ));
+          }, 3_000),
+        ),
+      ]);
+    } finally {
+      clearInterval(resolver);
+      if (timeout) clearTimeout(timeout);
+    }
+
+    assert.equal(pi.calls.length, 4, "verification retry should be capped after four dispatched attempts");
+    const stopEntry = deps.callLog.find((e: string) => e.startsWith("stopAuto:"));
+    assert.match(stopEntry ?? "", /requested retry 4 times without passing/);
+    const finalGraph = readGraph(runDir);
+    assert.equal(finalGraph.steps[0]?.status, "active", "failed verification must not reconcile the step complete");
+  });
+
+  it("persists custom verification retry budget across a session restart", async () => {
+    _resetPendingResolve();
+
+    const runDir = makeTmpDir();
+    const graph = makeGraph([makeStep({ id: "retry-step" })], "retry-restart");
+    writeGraph(runDir, graph);
+    writeFileSync(join(runDir, "DEFINITION.yaml"), stringify({
+      version: 1,
+      name: "retry-restart",
+      steps: [{
+        id: "retry-step",
+        name: "retry-step",
+        prompt: "Do retry-step",
+        produces: "retry-step/output.md",
+        verify: { policy: "shell-command", command: "exit 1" },
+      }],
+    }));
+
+    const ctx1 = makeMockCtx();
+    const pi1 = makeMockPi();
+    const s1 = makeLoopSession({
+      activeEngineId: "custom",
+      activeRunDir: runDir,
+      basePath: runDir,
+    });
+    const deps1 = makeMockDeps();
+    const resolver1 = setInterval(() => {
+      resolveAgentEnd({ messages: [{ role: "assistant" }] });
+      if (pi1.calls.length >= 2) {
+        s1.active = false;
+      }
+    }, 25);
+    let timeout1: NodeJS.Timeout | undefined;
+    try {
+      await Promise.race([
+        autoLoop(ctx1, pi1, s1, deps1),
+        new Promise((_, reject) =>
+          timeout1 = setTimeout(() => {
+            s1.active = false;
+            resolveAgentEnd({ messages: [{ role: "assistant" }] });
+            reject(new Error(
+              `first autoLoop did not pause after two retry attempts; calls=${pi1.calls.length}; log=${deps1.callLog.join(",")}`,
+            ));
+          }, 3_000),
+        ),
+      ]);
+    } finally {
+      clearInterval(resolver1);
+      if (timeout1) clearTimeout(timeout1);
+    }
+    assert.equal(pi1.calls.length, 2, "first session should consume two retry attempts");
+    assert.equal(
+      deps1.callLog.some((e: string) => e.startsWith("stopAuto:")),
+      false,
+      "first session should stop because the session deactivated, not because retry budget exhausted",
+    );
+
+    _resetPendingResolve();
+    const ctx2 = makeMockCtx();
+    const pi2 = makeMockPi();
+    const s2 = makeLoopSession({
+      activeEngineId: "custom",
+      activeRunDir: runDir,
+      basePath: runDir,
+    });
+    const deps2 = makeMockDeps({
+      stopAuto: async (_ctx, _pi, reason) => {
+        deps2.callLog.push(`stopAuto:${reason ?? "no-reason"}`);
+        s2.active = false;
+      },
+    });
+    const resolver2 = setInterval(() => {
+      resolveAgentEnd({ messages: [{ role: "assistant" }] });
+    }, 25);
+    let timeout2: NodeJS.Timeout | undefined;
+    try {
+      await Promise.race([
+        autoLoop(ctx2, pi2, s2, deps2),
+        new Promise((_, reject) =>
+          timeout2 = setTimeout(() => {
+            s2.active = false;
+            resolveAgentEnd({ messages: [{ role: "assistant" }] });
+            reject(new Error(
+              `second autoLoop did not stop after persisted retry exhaustion; calls=${pi2.calls.length}; log=${deps2.callLog.join(",")}`,
+            ));
+          }, 3_000),
+        ),
+      ]);
+    } finally {
+      clearInterval(resolver2);
+      if (timeout2) clearTimeout(timeout2);
+    }
+
+    assert.equal(pi2.calls.length, 2, "second session should exhaust after attempts 3 and 4");
+    const stopEntry = deps2.callLog.find((e: string) => e.startsWith("stopAuto:"));
+    assert.match(stopEntry ?? "", /requested retry 4 times without passing/);
+  });
+
   it("GRAPH.yaml step stays pending when session deactivates before reconcile", async () => {
     _resetPendingResolve();
 

package/src/resources/extensions/gsd/tests/db-access-guardrails.test.ts

@@ -50,6 +50,7 @@ test("doctor-heal.md contains DB access guardrail naming gsd_milestone_status",
   const prompt = readPrompt("doctor-heal");
   assert.match(prompt, /gsd_milestone_status/, "doctor-heal.md must name gsd_milestone_status as the DB inspection tool");
   assert.match(prompt, /Do NOT query.*\.gsd\/gsd\.db/i, "doctor-heal.md must prohibit direct DB queries");
+  assert.doesNotMatch(prompt, /\{\{milestoneId\}\}/, "doctor-heal.md must not declare unprovided milestoneId template variables");
 });
 
 test("forensics.md contains DB inspection guardrail", () => {

package/src/resources/extensions/gsd/tests/derive-state-crossval.test.ts

@@ -351,9 +351,9 @@ skills_used: []
       const dbState = await deriveStateFromDb(base);
 
       assertStatesEqual(dbState, fileState, 'E-blocked');
-      // With partial-dep fallback, circular deps no longer block — fallback picks first eligible slice
-      assert.deepStrictEqual(dbState.phase, 'planning', 'E-blocked: phase is planning (fallback picks a slice)');
-      assert.ok(dbState.activeSlice !== null, 'E-blocked: activeSlice is set via fallback');
+      assert.deepStrictEqual(dbState.phase, 'blocked', 'E-blocked: phase is blocked when no slice deps are satisfied');
+      assert.deepStrictEqual(dbState.activeSlice, null, 'E-blocked: no activeSlice is selected through unmet deps');
+      assert.ok(dbState.blockers.some(b => b.includes('No slice eligible')), 'E-blocked: blocker explains no eligible slice');
 
       closeDatabase();
     } finally {

package/src/resources/extensions/gsd/tests/derive-state-db-disk-reconcile.test.ts

@@ -112,6 +112,46 @@ async function main(): Promise<void> {
     cleanup(base);
   }
 
+  console.log("\n=== #4974: summary-only disk milestones keep parsed title ===");
+
+  {
+    const summaryOnlyBase = createFixtureBase();
+    const summaryOnlyDbPath = join(summaryOnlyBase, ".gsd", "gsd.db");
+
+    try {
+      openDatabase(summaryOnlyDbPath);
+      insertMilestone({ id: "M001", title: "M001: Existing DB Milestone", status: "complete", depends_on: [] });
+
+      writeFile(
+        summaryOnlyBase,
+        "milestones/M002/SUMMARY.md",
+        `---\nid: M002\nstatus: complete\n---\n\n# M002: Summary-Only Milestone\n\nComplete.`,
+      );
+
+      invalidateStateCache();
+      const state = await deriveStateFromDb(summaryOnlyBase);
+      const m002Entry = state.registry.find((m) => m.id === "M002");
+
+      assertTrue(
+        m002Entry !== undefined,
+        "M002 summary-only disk milestone should appear in state.registry (#4974)",
+      );
+      assertEq(
+        m002Entry?.title,
+        "Summary-Only Milestone",
+        "M002 summary-only disk milestone should use parsed SUMMARY title (#4974)",
+      );
+      assertEq(
+        m002Entry?.status,
+        "complete",
+        "M002 summary-only disk milestone should reconcile as complete (#4974)",
+      );
+    } finally {
+      closeDatabase();
+      cleanup(summaryOnlyBase);
+    }
+  }
+
   report();
 }
 

package/src/resources/extensions/gsd/tests/derive-state-db.test.ts

@@ -14,6 +14,7 @@ import {
   getAllMilestones,
   insertSlice,
   insertTask,
+  getSliceTasks,
   updateTaskStatus,
 } from '../gsd-db.ts';
 // ─── Fixture Helpers ───────────────────────────────────────────────────────
@@ -270,6 +271,93 @@ describe('derive-state-db', async () => {
     }
   });
 
+  test('derive-state-db: partial task rows reconcile missing plan tasks before summarizing', async (t) => {
+    const base = createFixtureBase();
+    t.after(() => {
+      closeDatabase();
+      cleanup(base);
+    });
+
+    const partialTaskPlan = `# S01: First Slice
+
+**Goal:** Test partial task DB reconciliation.
+**Demo:** Tests pass.
+
+## Tasks
+
+- [x] **T01: Existing Complete** \`est:10m\`
+  Already complete in DB.
+
+- [ ] **T02: Missing Pending** \`est:10m\`
+  Missing from DB but present in the plan.
+`;
+    writeFile(base, 'milestones/M001/M001-ROADMAP.md', ROADMAP_CONTENT);
+    writeFile(base, 'milestones/M001/slices/S01/S01-PLAN.md', partialTaskPlan);
+    writeFile(base, 'milestones/M001/slices/S01/tasks/.gitkeep', '');
+    writeFile(base, 'milestones/M001/slices/S01/tasks/T01-PLAN.md', '# T01 Plan');
+    writeFile(base, 'milestones/M001/slices/S01/tasks/T02-PLAN.md', '# T02 Plan');
+
+    openDatabase(':memory:');
+    insertMilestone({ id: 'M001', title: 'Test Milestone', status: 'active' });
+    insertSlice({ id: 'S01', milestoneId: 'M001', title: 'First Slice', status: 'active', risk: 'low', depends: [] });
+    insertTask({ id: 'T01', sliceId: 'S01', milestoneId: 'M001', title: 'Existing Complete', status: 'complete' });
+
+    invalidateStateCache();
+    const state = await deriveState(base);
+
+    const dbTasks = getSliceTasks('M001', 'S01');
+    assert.deepStrictEqual(dbTasks.length, 2, 'partial-task-db: missing T02 imported from plan');
+    assert.deepStrictEqual(dbTasks.find(t => t.id === 'T01')?.status, 'complete', 'partial-task-db: existing complete T01 preserved');
+    assert.deepStrictEqual(dbTasks.find(t => t.id === 'T02')?.status, 'pending', 'partial-task-db: missing T02 inserted pending');
+    assert.deepStrictEqual(state.phase, 'executing', 'partial-task-db: phase remains executing, not summarizing');
+    assert.deepStrictEqual(state.activeTask?.id, 'T02', 'partial-task-db: activeTask is missing plan task T02');
+    assert.deepStrictEqual(state.progress?.tasks, { done: 1, total: 2 }, 'partial-task-db: task progress includes reconciled plan task');
+  });
+
+  test('derive-state-db: empty DB disk import preserves milestone completion and dependencies', async (t) => {
+    const base = createFixtureBase();
+    t.after(() => {
+      closeDatabase();
+      cleanup(base);
+    });
+
+    writeFile(base, 'milestones/M001/M001-ROADMAP.md', `# M001: Foundation
+
+**Vision:** Foundation.
+
+## Slices
+
+- [x] **S01: Done** \`risk:low\` \`depends:[]\`
+  > After this: Done.
+`);
+    writeFile(base, 'milestones/M001/M001-VALIDATION.md', '---\nverdict: pass\nremediation_round: 0\n---\n\nPassed.');
+    writeFile(base, 'milestones/M001/M001-SUMMARY.md', '# M001 Summary\n\nDone.');
+    writeFile(base, 'milestones/M002/M002-CONTEXT.md', '---\ndepends_on:\n  - M001\n---\n\n# M002: Dependent\n');
+    writeFile(base, 'milestones/M002/M002-ROADMAP.md', `# M002: Dependent
+
+**Vision:** Active work.
+
+## Slices
+
+- [ ] **S01: Work** \`risk:low\` \`depends:[]\`
+  > After this: Done.
+`);
+    writeFile(base, 'milestones/M003/M003-CONTEXT.md', '---\ndepends_on:\n  - M002\n---\n\n# M003: Blocked\n');
+
+    openDatabase(':memory:');
+
+    invalidateStateCache();
+    const state = await deriveState(base);
+
+    const milestones = getAllMilestones();
+    assert.deepStrictEqual(milestones.find(m => m.id === 'M001')?.status, 'complete', 'disk-import: terminal summary imports M001 as complete');
+    assert.deepStrictEqual(milestones.find(m => m.id === 'M002')?.depends_on, ['M001'], 'disk-import: M002 depends_on imported from CONTEXT');
+    assert.deepStrictEqual(milestones.find(m => m.id === 'M003')?.depends_on, ['M002'], 'disk-import: M003 depends_on imported from CONTEXT');
+    assert.deepStrictEqual(state.activeMilestone?.id, 'M002', 'disk-import: M002 is active because M001 completion satisfies dependency');
+    assert.deepStrictEqual(state.registry.find(e => e.id === 'M001')?.status, 'complete', 'disk-import: M001 registry status is complete');
+    assert.deepStrictEqual(state.registry.find(e => e.id === 'M003')?.status, 'pending', 'disk-import: M003 stays pending on unmet M002 dependency');
+  });
+
   // ─── Test 5: Requirements counting from disk (DB no longer used for content) ─
   test('derive-state-db: requirements from disk content', async () => {
     const base = createFixtureBase();
@@ -616,10 +704,10 @@ describe('derive-state-db', async () => {
       invalidateStateCache();
       const dbState = await deriveStateFromDb(base);
 
-      // With partial-dep fallback, circular deps no longer block — fallback picks first eligible slice
-      assert.deepStrictEqual(dbState.phase, 'planning', 'blocked-db: phase is planning (fallback picks a slice)');
+      assert.deepStrictEqual(dbState.phase, 'blocked', 'blocked-db: phase is blocked when no slice deps are satisfied');
       assert.deepStrictEqual(dbState.phase, fileState.phase, 'blocked-db: phase matches filesystem');
-      assert.ok(dbState.activeSlice !== null, 'blocked-db: activeSlice is set via fallback');
+      assert.deepStrictEqual(dbState.activeSlice, null, 'blocked-db: no activeSlice is selected through unmet deps');
+      assert.ok(dbState.blockers.some(b => b.includes('No slice eligible')), 'blocked-db: blocker explains no eligible slice');
 
       closeDatabase();
     } finally {

package/src/resources/extensions/gsd/tests/derive-state.test.ts

@@ -431,7 +431,7 @@ Continue from step 2.
       cleanup(base1);
     }
 
-    // Case B: S01 depends on nonexistent S99 → truly blocked
+    // Case B: S01 depends on nonexistent S99 -> no slice is eligible
     const base2 = createFixtureBase();
     try {
       writeRoadmap(base2, 'M001', `# M001: Test Milestone
@@ -446,9 +446,9 @@ Continue from step 2.
 
       const state2 = await deriveState(base2);
 
-      // With partial-dep fallback, S01 is picked despite unmet dep on S99
-      assert.deepStrictEqual(state2.phase, 'planning', 'blocked-B: phase is planning (fallback picks S01)');
-      assert.deepStrictEqual(state2.activeSlice?.id, 'S01', 'blocked-B: activeSlice is S01 via fallback');
+      assert.deepStrictEqual(state2.phase, 'blocked', 'blocked-B: phase is blocked when dependency is unsatisfied');
+      assert.deepStrictEqual(state2.activeSlice, null, 'blocked-B: no activeSlice selected through unmet deps');
+      assert.ok(state2.blockers.some(b => b.includes('No slice eligible')), 'blocked-B: blocker explains no eligible slice');
     } finally {
       cleanup(base2);
     }

package/src/resources/extensions/gsd/tests/discuss-slice-structured-questions.test.ts

@@ -11,6 +11,7 @@ import { describe, it } from 'node:test'
 import assert from 'node:assert/strict'
 import { readFileSync } from 'node:fs'
 import { resolve } from 'node:path'
+import { extractSourceRegion } from "./test-helpers.ts";
 
 const template = readFileSync(
   resolve(process.cwd(), 'src', 'resources', 'extensions', 'gsd', 'prompts', 'guided-discuss-slice.md'),
@@ -37,7 +38,7 @@ describe('discuss-slice structuredQuestionsAvailable template variable', () => {
     const falseIdx = template.indexOf('`{{structuredQuestionsAvailable}}` is `false`')
     assert.ok(falseIdx !== -1)
 
-    const afterFalse = template.slice(falseIdx, falseIdx + 300)
+    const afterFalse = extractSourceRegion(template, '`{{structuredQuestionsAvailable}}` is `false`')
     assert.ok(
       afterFalse.includes('plain text'),
       'when structuredQuestionsAvailable is false, questions should be in plain text',

package/src/resources/extensions/gsd/tests/discuss-tool-scope-leak.test.ts

@@ -22,6 +22,7 @@ import { join, dirname } from "node:path";
 import { fileURLToPath } from "node:url";
 
 import { DISCUSS_TOOLS_ALLOWLIST } from "../constants.ts";
+import { extractSourceRegion } from "./test-helpers.ts";
 
 const __dirname = dirname(fileURLToPath(import.meta.url));
 const guidedFlowSource = readFileSync(join(__dirname, "..", "guided-flow.ts"), "utf-8");
@@ -58,7 +59,7 @@ describe("#3616 — discuss tool scoping must not leak across sessions", () => {
 		);
 		const newSessionStart = agentSessionSource.indexOf("async newSession(options?:");
 		assert.ok(newSessionStart >= 0, "should find newSession");
-		const body = agentSessionSource.slice(newSessionStart, newSessionStart + 3000);
+		const body = extractSourceRegion(agentSessionSource, "async newSession(options?:");
 
 		// Both branches (cwd-changed and cwd-unchanged) must include extension tools
 		assert.ok(

package/src/resources/extensions/gsd/tests/dispatch-complete-milestone-guard.test.ts

@@ -47,21 +47,26 @@ describe("completing-milestone dispatch guard (#4324)", () => {
     assert.ok(dispatchIdx > -1, "complete-milestone dispatch should exist after the skip guard");
   });
 
-  test("imports updateMilestoneStatus for DB reconciliation", () => {
-    assert.match(source, /import\s*\{[^}]*updateMilestoneStatus[^}]*\}\s*from\s*["']\.\/gsd-db/);
-  });
-
-  test("reconciles DB when SUMMARY exists on disk but DB is out of sync", () => {
+  test("classifies SUMMARY outcome and conditionally reconciles DB (#4658)", () => {
     const phaseCheck = source.indexOf('phase !== "completing-milestone"');
     // The SUMMARY-exists reconciliation guard must appear in this rule
     const summaryGuard = source.indexOf('resolveMilestoneFile(basePath, mid, "SUMMARY")', phaseCheck);
     assert.ok(summaryGuard > -1, "SUMMARY file check should exist in the completing-milestone rule");
 
+    const classifyCall = source.indexOf("classifyMilestoneSummaryContent", summaryGuard);
+    assert.ok(classifyCall > -1, "SUMMARY mismatch handling should classify summary content");
+    const dbGateBeforeClassify = source.indexOf("existingSummary && isDbAvailable()", summaryGuard);
+    assert.ok(
+      dbGateBeforeClassify === -1 || dbGateBeforeClassify > classifyCall,
+      "SUMMARY classification must not be gated on DB availability",
+    );
+
     const reconcileCall = source.indexOf('updateMilestoneStatus(mid, "complete"', summaryGuard);
-    assert.ok(reconcileCall > -1, "updateMilestoneStatus reconciliation should follow the SUMMARY check");
+    assert.ok(reconcileCall > -1, "successful SUMMARY should reconcile DB to complete");
 
-    // Reconciliation must happen before the dispatch action
-    const dispatchIdx = source.indexOf('unitType: "complete-milestone"', reconcileCall);
-    assert.ok(dispatchIdx > -1, "reconciliation should appear before the dispatch action");
+    const stopAction = source.indexOf('action: "stop"', summaryGuard);
+    assert.ok(stopAction > -1, "SUMMARY mismatch should return stop action");
+    const warningLevel = source.indexOf('level: "warning"', summaryGuard);
+    assert.ok(warningLevel > -1, "SUMMARY mismatch should be warning-level stop (pauses auto-mode)");
   });
 });