gsd-pi 2.76.0 → 2.77.0-dev.1594b263e

package/src/resources/extensions/gsd/tests/journal-integration.test.ts

@@ -21,6 +21,13 @@ import type { IterationContext, LoopState, PreDispatchData, IterationData } from
 import type { SessionLockStatus } from "../session-lock.js";
 import { runDispatch, runUnitPhase, runPreDispatch, runFinalize } from "../auto/phases.js";
 import { readUnitRuntimeRecord } from "../unit-runtime.js";
+import {
+  closeDatabase,
+  insertMilestone,
+  insertSlice,
+  insertTask,
+  openDatabase,
+} from "../gsd-db.js";
 
 // ─── Helpers ─────────────────────────────────────────────────────────────────
 
@@ -77,6 +84,8 @@ function makeMockDeps(
     autoWorktreeBranch: () => "auto/M001",
     resolveMilestoneFile: () => null,
     reconcileMergeState: () => "clean",
+    preflightCleanRoot: () => ({ stashPushed: false, summary: "" }),
+    postflightPopStash: () => {},
     getLedger: () => ({ units: [] }),
     getProjectTotals: () => ({ cost: 0 }),
     formatCost: (c: number) => `$${c.toFixed(2)}`,
@@ -642,6 +651,63 @@ test("terminal event is emitted on blocked state", async () => {
   assert.deepEqual((terminalEvents[0].data as any).blockers, ["Missing API key"]);
 });
 
+test("#4671: plan-v2 missing CONTEXT.md reaches dispatch recovery instead of pausing", async () => {
+  const basePath = mkdtempSync(join(tmpdir(), "gsd-4671-predispatch-"));
+  mkdirSync(join(basePath, ".gsd", "milestones", "M001", "slices", "S01", "tasks"), { recursive: true });
+  openDatabase(join(basePath, ".gsd", "gsd.db"));
+  try {
+    insertMilestone({ id: "M001", title: "Test", status: "active" });
+    insertSlice({
+      id: "S01",
+      milestoneId: "M001",
+      title: "Slice 1",
+      status: "in_progress",
+      sequence: 1,
+    });
+    insertTask({
+      id: "T01",
+      milestoneId: "M001",
+      sliceId: "S01",
+      title: "Task 1",
+      status: "pending",
+      keyFiles: ["src/task.ts"],
+      sequence: 1,
+    });
+
+    let pauseCalls = 0;
+    const capture = createEventCapture();
+    const deps = makeMockDeps(capture, {
+      pauseAuto: async () => { pauseCalls++; },
+      deriveState: async () => ({
+        phase: "executing",
+        activeMilestone: { id: "M001", title: "Test", status: "active" },
+        activeSlice: { id: "S01", title: "Slice 1" },
+        activeTask: { id: "T01", title: "Task 1" },
+        registry: [{ id: "M001", status: "active" }],
+        blockers: [],
+        recentDecisions: [],
+        nextAction: "dispatch",
+      }) as any,
+    });
+    const ic = makeIC(deps, {
+      prefs: { uok: { plan_v2: { enabled: true } } } as any,
+    });
+    ic.s.basePath = basePath;
+
+    const result = await runPreDispatch(ic, {
+      recentUnits: [],
+      stuckRecoveryAttempts: 0,
+      consecutiveFinalizeTimeouts: 0,
+    });
+
+    assert.equal(result.action, "next");
+    assert.equal(pauseCalls, 0, "missing CONTEXT.md should be handled by dispatch recovery, not plan gate pause");
+  } finally {
+    closeDatabase();
+    rmSync(basePath, { recursive: true, force: true });
+  }
+});
+
 test("milestone-transition event is emitted when milestone changes", async () => {
   const capture = createEventCapture();
   const deps = makeMockDeps(capture, {
@@ -837,3 +903,38 @@ test("runFinalize pauses and emits unit-end when pre-verification times out", as
   assert.equal((endEvents[0].data as any).artifactVerified, false);
   assert.equal((endEvents[0].data as any).finalizeStage, "pre");
 });
+
+test("transient session-failed cancellations pause instead of hard-stopping", async () => {
+  const capture = createEventCapture();
+  const { resolveAgentEndCancelled, _resetPendingResolve } = await import("../auto-loop.js");
+  _resetPendingResolve();
+
+  const deps = makeMockDeps(capture);
+  const ic = makeIC(deps);
+  const iterData: IterationData = {
+    unitType: "execute-task",
+    unitId: "M001/S01/T02",
+    prompt: "do more stuff",
+    finalPrompt: "do more stuff",
+    pauseAfterUatDispatch: false,
+    state: { phase: "executing", activeMilestone: { id: "M001" }, activeSlice: { id: "S01" }, registry: [], blockers: [] } as any,
+    mid: "M001",
+    midTitle: "Test",
+    isRetry: false,
+    previousTier: undefined,
+  };
+  const loopState: LoopState = { recentUnits: [{ key: "execute-task/M001/S01/T02" }], stuckRecoveryAttempts: 0, consecutiveFinalizeTimeouts: 0 };
+
+  const unitPromise = runUnitPhase(ic, iterData, loopState);
+  await new Promise(r => setTimeout(r, 50));
+
+  resolveAgentEndCancelled({ message: "Session creation failed: temporary bootstrap overload", category: "session-failed", isTransient: true });
+
+  const result = await unitPromise;
+  assert.equal(result.action, "break");
+  assert.equal((result as any).reason, "session-timeout");
+
+  const entry = loopState.recentUnits[loopState.recentUnits.length - 1];
+  assert.ok(entry.error, "window entry must have error set");
+  assert.ok(entry.error!.startsWith("session-failed:"), "error must preserve the session-failed category");
+});

package/src/resources/extensions/gsd/tests/key-manager.test.ts

@@ -141,6 +141,15 @@ test("PROVIDER_REGISTRY includes all major LLM providers", () => {
   assert.ok(ids.includes("openai"));
   assert.ok(ids.includes("google"));
   assert.ok(ids.includes("groq"));
+  assert.ok(ids.includes("minimax"));
+  assert.ok(ids.includes("minimax-cn"));
+});
+
+test("PROVIDER_REGISTRY includes claude-code as a first-class LLM provider (#4541)", () => {
+  const entry = PROVIDER_REGISTRY.find((p) => p.id === "claude-code");
+  assert.ok(entry, "claude-code must be in PROVIDER_REGISTRY");
+  assert.equal(entry!.category, "llm");
+  assert.ok(entry!.hasOAuth, "claude-code uses OAuth (CLI auth)");
 });
 
 test("PROVIDER_REGISTRY includes all tool/search providers", () => {

package/src/resources/extensions/gsd/tests/knowledge.test.ts

@@ -15,7 +15,7 @@ import { mkdtempSync, mkdirSync, writeFileSync, readFileSync, rmSync, realpathSy
 import { join } from 'node:path';
 import { tmpdir } from 'node:os';
 import { GSD_ROOT_FILES, resolveGsdRootFile } from '../paths.ts';
-import { inlineGsdRootFile } from '../auto-prompts.ts';
+import { inlineGsdRootFile, inlineKnowledgeBudgeted } from '../auto-prompts.ts';
 import { appendKnowledge } from '../files.ts';
 import { loadKnowledgeBlock } from '../bootstrap/system-context.ts';
 
@@ -248,3 +248,95 @@ test('loadKnowledgeBlock: reports globalSizeKb above 4KB threshold', () => {
 
   rmSync(tmp, { recursive: true, force: true });
 });
+
+// ─── inlineKnowledgeBudgeted — issue #4719 ─────────────────────────────────
+// Milestone-phase prompts must not inject the full KNOWLEDGE.md. The budgeted
+// helper scopes by milestone-level keywords and caps the injected size.
+
+test('inlineKnowledgeBudgeted: returns scoped H3 entries for single-H2 file', async () => {
+  const tmp = realpathSync(mkdtempSync(join(tmpdir(), 'gsd-knowledge-')));
+  const gsdDir = join(tmp, '.gsd');
+  mkdirSync(gsdDir, { recursive: true });
+
+  const content = `# 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.
+`;
+  writeFileSync(join(gsdDir, 'KNOWLEDGE.md'), content);
+
+  const result = await inlineKnowledgeBudgeted(tmp, ['database']);
+  assert.ok(result !== null, 'should return content');
+  assert.ok(result!.includes('Database: prepared statements'), 'includes matching H3');
+  assert.ok(!result!.includes('API: versioned paths'), 'excludes non-matching H3');
+
+  rmSync(tmp, { recursive: true, force: true });
+});
+
+test('inlineKnowledgeBudgeted: caps payload below budget for large files', async () => {
+  const tmp = realpathSync(mkdtempSync(join(tmpdir(), 'gsd-knowledge-')));
+  const gsdDir = join(tmp, '.gsd');
+  mkdirSync(gsdDir, { recursive: true });
+
+  // Build a 200KB KNOWLEDGE with 500 H3 entries all matching 'shared'
+  const entries = Array.from({ length: 500 }, (_, i) =>
+    `### Entry ${i}: shared topic\n${'filler text '.repeat(30)}\n`,
+  ).join('\n');
+  const content = `# Project Knowledge\n\n## Patterns\n\n${entries}`;
+  writeFileSync(join(gsdDir, 'KNOWLEDGE.md'), content);
+
+  const BUDGET_CHARS = 30_000;
+  const result = await inlineKnowledgeBudgeted(tmp, ['shared'], { maxChars: BUDGET_CHARS });
+  assert.ok(result !== null, 'should return content');
+  // Allow some overhead for header formatting, but must stay close to budget
+  assert.ok(
+    result!.length <= BUDGET_CHARS + 500,
+    `payload ${result!.length} chars should be <= budget ${BUDGET_CHARS} (+overhead)`,
+  );
+  // Far smaller than the raw file
+  assert.ok(
+    result!.length < content.length / 4,
+    `payload should be much smaller than full content (${content.length} chars)`,
+  );
+  assert.match(
+    result!,
+    /\[\.\.\.truncated \d+ chars; rerun with narrower scope if needed\]/,
+    'should include truncation note when budget is exceeded',
+  );
+
+  rmSync(tmp, { recursive: true, force: true });
+});
+
+test('inlineKnowledgeBudgeted: returns null when no KNOWLEDGE.md exists', async () => {
+  const tmp = realpathSync(mkdtempSync(join(tmpdir(), 'gsd-knowledge-')));
+  const gsdDir = join(tmp, '.gsd');
+  mkdirSync(gsdDir, { recursive: true });
+
+  const result = await inlineKnowledgeBudgeted(tmp, ['database']);
+  assert.strictEqual(result, null);
+
+  rmSync(tmp, { recursive: true, force: true });
+});
+
+test('inlineKnowledgeBudgeted: returns null when no entries match', async () => {
+  const tmp = realpathSync(mkdtempSync(join(tmpdir(), 'gsd-knowledge-')));
+  const gsdDir = join(tmp, '.gsd');
+  mkdirSync(gsdDir, { recursive: true });
+  writeFileSync(
+    join(gsdDir, 'KNOWLEDGE.md'),
+    '# Project Knowledge\n\n## Patterns\n\n### Database\nuse it\n',
+  );
+
+  const result = await inlineKnowledgeBudgeted(tmp, ['nonexistent']);
+  assert.strictEqual(result, null);
+
+  rmSync(tmp, { recursive: true, force: true });
+});

package/src/resources/extensions/gsd/tests/load-memory-block.test.ts

@@ -0,0 +1,36 @@
+// GSD-2 — loadMemoryBlock tests (ADR-013 step 4 auto-injection parity)
+import { test } from 'node:test';
+import assert from 'node:assert/strict';
+
+import { closeDatabase, openDatabase } from '../gsd-db.ts';
+import { createMemory } from '../memory-store.ts';
+import { loadMemoryBlock } from '../bootstrap/system-context.ts';
+
+// ─── Success path: critical memories surface in the labeled block ──────────
+
+test('loadMemoryBlock: renders MEMORY block when critical memories exist', async () => {
+  openDatabase(':memory:');
+  try {
+    const id = createMemory({
+      category: 'architecture',
+      content: 'Use the memories table as the single source of truth for decisions.',
+      confidence: 0.95,
+    });
+    assert.ok(id, 'createMemory should seed a memory');
+
+    const block = await loadMemoryBlock('');
+    assert.ok(block.length > 0, 'block should be non-empty when critical memories exist');
+    assert.match(block, /\[MEMORY — Critical and prompt-relevant memories/);
+    assert.match(block, /memories table as the single source of truth/);
+  } finally {
+    closeDatabase();
+  }
+});
+
+// ─── Failure / degraded path: no DB → returns "" without throwing ───────────
+
+test('loadMemoryBlock: returns empty string when no DB is available', async () => {
+  closeDatabase();
+  const block = await loadMemoryBlock('anything');
+  assert.equal(block, '', 'no DB → empty block (graceful degradation)');
+});

package/src/resources/extensions/gsd/tests/mcp-client-security.test.ts

@@ -0,0 +1,47 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+
+import {
+  _buildMcpChildEnvForTest,
+  _buildMcpTrustConfirmOptionsForTest,
+} from "../../mcp-client/index.ts";
+
+// Note: four source-grep tests that scanned `mcp-client/index.ts` for
+// Map<> shapes, catch-block structure, and closeAll body were removed
+// under #4827. They encoded implementation shape rather than behaviour —
+// any refactor (extracted helper, different Map key type, rearranged
+// cleanup order) broke the greps without a real regression. Runtime
+// coverage of connectServer/closeAll with a mocked failing transport
+// is tracked as a follow-up.
+
+test("MCP stdio child env only includes safe inherited keys plus explicit config env", () => {
+  const previousSecret = process.env.SECRET_MCP_TEST_TOKEN;
+  const previousPath = process.env.PATH;
+  try {
+    process.env.SECRET_MCP_TEST_TOKEN = "should-not-leak";
+    process.env.PATH = "/usr/bin";
+
+    const env = _buildMcpChildEnvForTest({
+      EXPLICIT_TOKEN: "${SECRET_MCP_TEST_TOKEN}",
+      PLAIN_VALUE: "ok",
+    });
+
+    assert.equal(env.PATH, "/usr/bin");
+    assert.equal(env.SECRET_MCP_TEST_TOKEN, undefined);
+    assert.equal(env.EXPLICIT_TOKEN, "should-not-leak");
+    assert.equal(env.PLAIN_VALUE, "ok");
+  } finally {
+    if (previousSecret === undefined) delete process.env.SECRET_MCP_TEST_TOKEN;
+    else process.env.SECRET_MCP_TEST_TOKEN = previousSecret;
+    if (previousPath === undefined) delete process.env.PATH;
+    else process.env.PATH = previousPath;
+  }
+});
+
+test("MCP stdio trust confirmation is abort-aware", () => {
+  const controller = new AbortController();
+  const options = _buildMcpTrustConfirmOptionsForTest(controller.signal);
+
+  assert.equal(options.timeout, 120_000);
+  assert.equal(options.signal, controller.signal);
+});

package/src/resources/extensions/gsd/tests/md-importer.test.ts

@@ -363,7 +363,7 @@ test('md-importer: schema v1→v2 migration', () => {
   openDatabase(':memory:');
   const adapter = _getAdapter();
   const version = adapter?.prepare('SELECT MAX(version) as v FROM schema_version').get();
-  assert.deepStrictEqual(version?.v, 20, 'new DB should be at schema version 20');
+  assert.deepStrictEqual(version?.v, 22, 'new DB should be at schema version 22');
 
   // Artifacts table should exist
   const tableCheck = adapter?.prepare("SELECT count(*) as c FROM sqlite_master WHERE type='table' AND name='artifacts'").get();

package/src/resources/extensions/gsd/tests/memory-extractor.test.ts

@@ -1,4 +1,4 @@
-import { parseMemoryResponse, _resetExtractionState, buildMemoryLLMCall } from '../memory-extractor.ts';
+import { parseMemoryResponse, buildMemoryLLMCall } from '../memory-extractor.ts';
 import {
   openDatabase,
   closeDatabase,
@@ -159,16 +159,6 @@ test('integration: mixed action lifecycle', () => {
   closeDatabase();
 });
 
-// ═══════════════════════════════════════════════════════════════════════════
-// memory-extractor: _resetExtractionState
-// ═══════════════════════════════════════════════════════════════════════════
-
-test('memory-extractor: reset extraction state', () => {
-  // Just verify it doesn't throw
-  _resetExtractionState();
-  assert.ok(true, '_resetExtractionState should not throw');
-});
-
 // ═══════════════════════════════════════════════════════════════════════════
 // memory-extractor: buildMemoryLLMCall resolves OAuth API key via modelRegistry
 // Regression test for #2959 — OAuth users had broken memory extraction
@@ -200,8 +190,8 @@ test('memory-extractor: buildMemoryLLMCall resolves API key from modelRegistry f
   assert.ok(llmCallFn !== null, 'buildMemoryLLMCall should return a function when models are available');
 
   // The function should have resolved the API key eagerly via modelRegistry.getApiKey.
-  // Give the async getApiKey a tick to resolve.
-  await new Promise(resolve => setTimeout(resolve, 50));
+  // Await the exposed promise deterministically instead of polling via setTimeout.
+  await llmCallFn!.apiKeyReady;
   assert.ok(getApiKeyCalled, 'buildMemoryLLMCall must call modelRegistry.getApiKey() to resolve OAuth tokens');
 });
 
@@ -246,8 +236,8 @@ test('memory-extractor: buildMemoryLLMCall prefers haiku model', async () => {
   const llmCallFn = buildMemoryLLMCall(ctx);
   assert.ok(llmCallFn !== null, 'should return a function');
 
-  // Wait for the async getApiKey to resolve
-  await new Promise(resolve => setTimeout(resolve, 50));
+  // Await the exposed API-key-ready promise deterministically.
+  await llmCallFn!.apiKeyReady;
   assert.strictEqual(resolvedModelId, 'claude-3-5-haiku-20241022',
     'should resolve API key for haiku model, not sonnet');
 });

package/src/resources/extensions/gsd/tests/memory-pressure-stuck-state.test.ts

@@ -51,4 +51,16 @@ describe("stuck detection persistence (#3704)", () => {
   test("stuck state file path uses runtime directory", () => {
     assert.match(loopSource, /stuck-state\.json/);
   });
+
+  test("saveStuckState called in standard dev path as well as custom engine path (#4382)", () => {
+    // Count all call-sites of saveStuckState (excluding the function definition itself).
+    // After the fix, both the custom-engine path and the standard dev path must each
+    // call saveStuckState so stuckRecoveryAttempts survives session restarts.
+    const callMatches = loopSource.match(/saveStuckState\(s\.basePath,\s*loopState\)/g) ?? [];
+    assert.ok(
+      callMatches.length >= 2,
+      `saveStuckState must be called in both the custom-engine path and the standard dev path ` +
+      `(found ${callMatches.length} call(s) — standard path is missing its call, #4382)`,
+    );
+  });
 });

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

@@ -323,9 +323,9 @@ test('memory-store: schema includes memories table', () => {
   const viewCount = adapter.prepare('SELECT count(*) as cnt FROM active_memories').get();
   assert.deepStrictEqual(viewCount?.['cnt'], 0, 'active_memories view should exist');
 
-  // Verify schema version is 20 (memory Phase 4 relations included)
+  // Verify schema version is 22 (v22 quality_gates DDL fix included)
   const version = adapter.prepare('SELECT MAX(version) as v FROM schema_version').get();
-  assert.deepStrictEqual(version?.['v'], 20, 'schema version should be 20');
+  assert.deepStrictEqual(version?.["v"], 22, 'schema version should be 22');
 
   closeDatabase();
 });