@opengsd/gsd-pi 1.2.0-dev.955e4da0 → 1.2.0-dev.a6376d75

package/src/resources/extensions/gsd/tests/oauth-api-model-routing.test.ts

@@ -0,0 +1,167 @@
+/**
+ * OAuth/subscription vs pay-per-token API routing for bare model IDs.
+ *
+ * When the same model ID (e.g. gpt-5.5) exists on multiple providers, resolveModelId
+ * must prefer subscription/OAuth routes over platform API keys.
+ */
+
+import test from "node:test";
+import assert from "node:assert/strict";
+import { MODELS } from "../../../../../packages/pi-ai/dist/models.generated.js";
+import {
+  BARE_ID_SUBSCRIPTION_PROVIDER_PRECEDENCE,
+  resolveModelId,
+} from "../auto-model-selection.js";
+
+type ModelRef = { id: string; provider: string };
+
+function modelsForProviders(modelId: string, providers: string[]): ModelRef[] {
+  return providers.map((provider) => ({ id: modelId, provider }));
+}
+
+/** Table-driven precedence cases — one row per conflict shape, not per model ID. */
+const PRECEDENCE_CASES: Array<{
+  label: string;
+  modelId: string;
+  providers: string[];
+  expected: string;
+  currentProvider?: string;
+}> = [
+  {
+    label: "ChatGPT OAuth beats OpenAI API (gpt-5.5)",
+    modelId: "gpt-5.5",
+    providers: ["openai", "openai-codex"],
+    expected: "openai-codex",
+  },
+  {
+    label: "Codex beats Copilot beats OpenAI API (gpt-5.5 triple)",
+    modelId: "gpt-5.5",
+    providers: ["openai", "github-copilot", "openai-codex"],
+    expected: "openai-codex",
+  },
+  {
+    label: "Copilot OAuth beats OpenAI API when no Codex (gpt-5.5)",
+    modelId: "gpt-5.5",
+    providers: ["openai", "github-copilot"],
+    expected: "github-copilot",
+  },
+  {
+    label: "Copilot OAuth beats OpenAI API (gpt-5-mini)",
+    modelId: "gpt-5-mini",
+    providers: ["openai", "github-copilot"],
+    expected: "github-copilot",
+  },
+  {
+    label: "Anthropic beats Copilot for Claude (claude-sonnet-4-6)",
+    modelId: "claude-sonnet-4-6",
+    providers: ["anthropic", "github-copilot"],
+    expected: "anthropic",
+  },
+  {
+    label: "Copilot beats Google API for Gemini (gemini-2.5-pro)",
+    modelId: "gemini-2.5-pro",
+    providers: ["google", "github-copilot"],
+    expected: "github-copilot",
+  },
+  {
+    label: "Gemini CLI beats Google API (gemini-2.5-pro)",
+    modelId: "gemini-2.5-pro",
+    providers: ["google", "google-gemini-cli"],
+    expected: "google-gemini-cli",
+  },
+  {
+    label: "Gemini CLI beats Copilot beats Google API",
+    modelId: "gemini-2.5-pro",
+    providers: ["google", "github-copilot", "google-gemini-cli"],
+    expected: "google-gemini-cli",
+  },
+  {
+    label: "Session provider still wins (openai-codex explicit)",
+    modelId: "gpt-5.5",
+    providers: ["openai", "openai-codex", "github-copilot"],
+    expected: "github-copilot",
+    currentProvider: "github-copilot",
+  },
+  {
+    label: "claude-code session beats anthropic (#3772)",
+    modelId: "claude-sonnet-4-6",
+    providers: ["anthropic", "claude-code"],
+    expected: "claude-code",
+    currentProvider: "claude-code",
+  },
+];
+
+for (const row of PRECEDENCE_CASES) {
+  test(`resolveModelId precedence: ${row.label}`, () => {
+    const available = modelsForProviders(row.modelId, row.providers);
+    const result = resolveModelId(row.modelId, available, row.currentProvider);
+    assert.ok(result, `expected a match for ${row.modelId}`);
+    assert.equal(result.provider, row.expected);
+  });
+}
+
+test("BARE_ID_SUBSCRIPTION_PROVIDER_PRECEDENCE covers known OAuth/subscription providers", () => {
+  const expected = [
+    "openai-codex",
+    "google-gemini-cli",
+    "anthropic",
+    "github-copilot",
+    "google-antigravity",
+  ];
+  assert.deepEqual([...BARE_ID_SUBSCRIPTION_PROVIDER_PRECEDENCE], expected);
+});
+
+test("every catalog overlap with OAuth/subscription + API resolves to subscription route", () => {
+  const oauthProviders = new Set(["anthropic", "github-copilot", "openai-codex"]);
+  const subscriptionCli = new Set(["google-gemini-cli", "google-antigravity", "claude-code"]);
+  const apiPayPerToken = new Set(["openai", "google", "azure-openai-responses"]);
+
+  const byId = new Map<string, Set<string>>();
+  for (const [provider, models] of Object.entries(MODELS)) {
+    for (const id of Object.keys(models)) {
+      if (!byId.has(id)) byId.set(id, new Set());
+      byId.get(id)!.add(provider);
+    }
+  }
+
+  const failures: string[] = [];
+
+  for (const [modelId, providers] of byId) {
+    if (providers.size < 2) continue;
+
+    const hasApi = [...providers].some((p) => apiPayPerToken.has(p));
+    const subscriptionCandidates = [...providers].filter(
+      (p) => oauthProviders.has(p) || subscriptionCli.has(p),
+    );
+    if (!hasApi || subscriptionCandidates.length === 0) continue;
+
+    const testProviders = new Set<string>(subscriptionCandidates);
+    for (const p of providers) {
+      if (apiPayPerToken.has(p)) testProviders.add(p);
+    }
+
+    const available = modelsForProviders(modelId, [...testProviders]);
+    const result = resolveModelId(modelId, available, undefined);
+    if (!result) {
+      failures.push(`${modelId}: no resolution among ${[...testProviders].join(", ")}`);
+      continue;
+    }
+
+    const winnerIsSubscription =
+      oauthProviders.has(result.provider) ||
+      subscriptionCli.has(result.provider) ||
+      result.provider === "anthropic";
+
+    if (!winnerIsSubscription) {
+      failures.push(
+        `${modelId}: got ${result.provider}, expected subscription route among ${[...testProviders].join(", ")}`,
+      );
+    }
+  }
+
+  assert.equal(
+    failures.length,
+    0,
+    `catalog overlap routing failures:\n${failures.join("\n")}`,
+  );
+});

package/src/resources/extensions/gsd/tests/parallel-research-dispatch.test.ts

@@ -41,6 +41,7 @@ after(() => {
 
 const { resolveDispatch } = await import("../auto-dispatch.ts");
 const { buildParallelResearchSlicesPrompt } = await import("../auto-prompts.ts");
+const { openDatabase, closeDatabase, insertMilestone, insertSlice } = await import("../gsd-db.ts");
 
 type DispatchState = Parameters<typeof resolveDispatch>[0]["state"];
 
@@ -51,6 +52,20 @@ function writeRoadmap(
 ): void {
   const milestoneDir = join(base, ".gsd", "milestones", mid);
   mkdirSync(milestoneDir, { recursive: true });
+  // The dispatch rule reads slices from the DB (ADR-017), not the ROADMAP
+  // projection — seed both so disk-based artifact checks still resolve.
+  insertMilestone({ id: mid, title: "Parallel Research Milestone" });
+  let sequence = 1;
+  for (const s of slices) {
+    insertSlice({
+      id: s.id,
+      milestoneId: mid,
+      title: s.title,
+      status: s.done ? "complete" : "pending",
+      depends: s.depends ?? [],
+      sequence: sequence++,
+    });
+  }
   const lines = [
     `# ${mid}: Parallel Research Milestone`,
     "",
@@ -100,9 +115,12 @@ describe("parallel-research-slices dispatch rule", () => {
 
   beforeEach(() => {
     base = mkdtempSync(join(tmpdir(), "parallel-research-"));
+    mkdirSync(join(base, ".gsd"), { recursive: true });
+    openDatabase(join(base, ".gsd", "gsd.db"));
   });
 
   afterEach(() => {
+    closeDatabase();
     rmSync(base, { recursive: true, force: true });
   });
 

package/src/resources/extensions/gsd/tests/parallel-skill-prompt-integration.test.ts

@@ -29,8 +29,8 @@ import {
 
 const SKILL_NAME = "testskill";
 const COMPLETE_SLICE_SKILL_NAME = "complete-slice-policies";
-const SKILL_ACTIVATION_SUBSTRING = `Call Skill({ skill: '${SKILL_NAME}' })`;
-const COMPLETE_SLICE_SKILL_ACTIVATION_SUBSTRING = `Call Skill({ skill: '${COMPLETE_SLICE_SKILL_NAME}' })`;
+const SKILL_ACTIVATION_SUBSTRING = `Read the installed '${SKILL_NAME}' skill file from <available_skills>`;
+const COMPLETE_SLICE_SKILL_ACTIVATION_SUBSTRING = `Read the installed '${COMPLETE_SLICE_SKILL_NAME}' skill file from <available_skills>`;
 
 const tmpDirs: string[] = [];
 let savedCwd: string | undefined;

package/src/resources/extensions/gsd/tests/parsers-legacy-importers.test.ts

@@ -0,0 +1,143 @@
+// Structural invariant: parsers-legacy is banned from decision paths (ADR-017).
+//
+// The DB is the single source of truth; `.gsd/*.md` files are projections.
+// Dispatch/gate/completion code must read state via gsd-db queries (e.g.
+// getMilestoneSliceSummaries), never by parsing markdown projections.
+//
+// Two assertions:
+// 1. Decision-path modules must NOT import parsers-legacy (hard ban).
+// 2. Every other importer must be on the explicit allowlist below, each with
+//    a one-line justification. When this test fails, do not extend the
+//    allowlist for a decision path — add/extend a query in db/queries.ts and
+//    read the DB instead.
+
+import test from "node:test";
+import assert from "node:assert/strict";
+import { readFileSync, readdirSync } from "node:fs";
+import { join, relative } from "node:path";
+
+const extensionsDir = join(process.cwd(), "src/resources/extensions");
+
+// Modules that make dispatch/gate/completion decisions. Importing
+// parsers-legacy here is always a violation, allowlist or not.
+const BANNED_DECISION_PATHS = new Set([
+  "gsd/auto-direct-dispatch.ts",
+  "gsd/dispatch-guard.ts",
+  "gsd/auto-verification.ts",
+  "gsd/auto-dispatch.ts",
+  "gsd/auto-post-unit.ts",
+  "gsd/milestone-closeout.ts",
+  "gsd/auto/phases.ts",
+  "gsd/auto/pre-dispatch.ts",
+  "gsd/auto/dispatch.ts",
+  "gsd/auto/unit-phase.ts",
+  "gsd/auto/finalize.ts",
+  "gsd/auto/closeout.ts",
+  "gsd/auto/orchestrator.ts",
+  "gsd/auto/loop.ts",
+  "gsd/tools/complete-slice.ts",
+]);
+
+// Legitimate importers. Each entry carries its justification; anything not
+// listed here (and not under a tests/ directory) fails the test.
+const ALLOWED_IMPORTERS = new Set([
+  // migration/import: parses markdown to populate the DB
+  "gsd/md-importer.ts",
+  "gsd/workflow-migration.ts",
+  "gsd/migration-auto-check.ts",
+  // drift detection: compares markdown projection against DB by design
+  "gsd/state-reconciliation/drift/roadmap.ts",
+  // stale-render detection + render verification helpers over rendered output
+  "gsd/markdown-renderer.ts",
+  // pre-migration fallback: deriveState must work before the DB exists
+  "gsd/state.ts",
+  // explicit degraded-mode fallback when DB has no task rows (warns on use)
+  "gsd/reactive-graph.ts",
+  // recovery path: explicit pre-migration/DB-unavailable fallback branches
+  "gsd/auto-recovery.ts",
+  // diagnostics-only surfaces: report on projections, make no dispatch decisions
+  "gsd/doctor.ts",
+  // display/telemetry-only surfaces
+  "gsd/workspace-index.ts",
+  "gsd/visualizer-data.ts",
+  // prompt context text (display strings injected into unit prompts)
+  "gsd/auto-prompts.ts",
+  // cold-path maintenance command (branch cleanup messaging)
+  "gsd/commands-maintenance.ts",
+  // display-only GitHub issue/PR body sync
+  "github-sync/sync.ts",
+]);
+
+const IMPORT_RE = /from\s+["'][^"']*parsers-legacy(?:\.js)?["']|import\(\s*["'][^"']*parsers-legacy(?:\.js)?["']\s*\)|require\(\s*["'][^"']*parsers-legacy(?:\.js)?["']\s*\)/;
+
+function walkTsFiles(root: string): string[] {
+  const out: string[] = [];
+  const stack: string[] = [root];
+
+  while (stack.length > 0) {
+    const dir = stack.pop()!;
+    let entries;
+    try {
+      entries = readdirSync(dir, { withFileTypes: true });
+    } catch {
+      continue;
+    }
+    for (const ent of entries) {
+      const full = join(dir, ent.name);
+      if (ent.isDirectory()) {
+        if (ent.name === "tests" || ent.name === "node_modules") continue;
+        stack.push(full);
+        continue;
+      }
+      if (!ent.isFile()) continue;
+      if (!ent.name.endsWith(".ts")) continue;
+      if (ent.name.startsWith(".")) continue;
+      if (ent.name.endsWith(".test.ts")) continue;
+      out.push(full);
+    }
+  }
+  return out;
+}
+
+function findImporters(): string[] {
+  const importers: string[] = [];
+  for (const file of walkTsFiles(extensionsDir)) {
+    const rel = relative(extensionsDir, file).split("\\").join("/");
+    if (rel === "gsd/parsers-legacy.ts") continue; // the module itself
+    const content = readFileSync(file, "utf-8");
+    if (IMPORT_RE.test(content)) importers.push(rel);
+  }
+  return importers.sort();
+}
+
+test("decision-path modules do not import parsers-legacy (ADR-017)", () => {
+  const violations = findImporters().filter((rel) => BANNED_DECISION_PATHS.has(rel));
+  assert.deepEqual(
+    violations,
+    [],
+    `Decision-path modules must read the DB (db/queries.ts, e.g. getMilestoneSliceSummaries), ` +
+      `not parse .gsd/*.md projections. Violations:\n  ${violations.join("\n  ")}`,
+  );
+});
+
+test("every parsers-legacy importer is on the explicit allowlist", () => {
+  const unexpected = findImporters().filter((rel) => !ALLOWED_IMPORTERS.has(rel));
+  assert.deepEqual(
+    unexpected,
+    [],
+    `New parsers-legacy importer(s) detected:\n  ${unexpected.join("\n  ")}\n` +
+      `If this is migration/drift/display-only code, add it to ALLOWED_IMPORTERS ` +
+      `with a one-line justification. If it makes dispatch/gate/completion ` +
+      `decisions, read the DB instead (db/queries.ts).`,
+  );
+});
+
+test("allowlist has no stale entries", () => {
+  const importers = new Set(findImporters());
+  const stale = [...ALLOWED_IMPORTERS].filter((rel) => !importers.has(rel));
+  assert.deepEqual(
+    stale,
+    [],
+    `Allowlist entries no longer import parsers-legacy — remove them:\n  ${stale.join("\n  ")}`,
+  );
+});

package/src/resources/extensions/gsd/tests/phases-merge-error-stops-auto.test.ts

@@ -10,7 +10,7 @@
  */
 
 import { createTestContext } from "./test-helpers.ts";
-import { runPreDispatch } from "../auto/phases.ts";
+import { runPreDispatch } from "../auto/pre-dispatch.ts";
 
 const { assertTrue, report } = createTestContext();
 

package/src/resources/extensions/gsd/tests/phases-terminal-complete-idempotent.test.ts

@@ -0,0 +1,242 @@
+/**
+ * phases-terminal-complete-idempotent.test.ts — Regression test for the
+ * milestone-completion double-closeout guard in `runPreDispatch`.
+ *
+ * When `runPreDispatch` observes that the active milestone (or the last one
+ * the session was working on) has already been closed by another session,
+ * the loop must exit cleanly with `{ action: "break", reason:
+ * "milestone-complete" }` and must NOT replay merge, desktop / cmux
+ * notifications, unit closeout, or `stopAuto`.
+ *
+ * There are two `deriveState` shapes the guard must cover, and both are
+ * exercised below because the canonical one is easy to miss:
+ *
+ *   1. `state.phase === "complete"` with `activeMilestone` set, so the
+ *      loop computes a non-null `mid` from `state.activeMilestone.id`.
+ *
+ *   2. `state.phase === "complete"` with `activeMilestone: null` — the
+ *      canonical "all milestones complete" return from `deriveState`
+ *      (state.ts:613, state.ts:1293). `mid` is undefined here, so the
+ *      guard must consult `s.currentMilestoneId` (the milestone this
+ *      session was working on) instead. Without coverage for this case,
+ *      a guard that only inspects `mid` is unreachable in production and
+ *      the loop replays `_runMilestoneMergeOnceWithStashRestore`,
+ *      `sendDesktopNotification("All milestones complete!")`,
+ *      `logCmuxEvent`, and `stopAuto` — exactly the duplicate side
+ *      effects this fix exists to prevent.
+ *
+ * Both fire-paths of the guard (`completionStopInProgress` and a
+ * DB-already-closed milestone) are exercised against each shape.
+ */
+
+import { createTestContext } from "./test-helpers.ts";
+import { runPreDispatch } from "../auto/pre-dispatch.ts";
+import {
+  openDatabase,
+  closeDatabase,
+  insertMilestone,
+  isDbAvailable,
+} from "../gsd-db.ts";
+
+const { assertTrue, report } = createTestContext();
+
+type SideEffect = string;
+
+interface ScenarioOverrides {
+  completionStopInProgress: boolean;
+  sideEffects: SideEffect[];
+  notifications: Array<{ message: string; level?: string }>;
+  // When `null`, deriveState returns activeMilestone: null (canonical
+  // all-complete path). Otherwise, returns an activeMilestone with this id.
+  activeMilestone: { id: string; title: string } | null;
+}
+
+function makeIterationContext(overrides: ScenarioOverrides): any {
+  const basePath = "/tmp/gsd-test-terminal-complete";
+  const recordSideEffect = (label: string) => {
+    overrides.sideEffects.push(label);
+  };
+  return {
+    ctx: {
+      ui: {
+        notify(message: string, level?: string) {
+          overrides.notifications.push({ message, level });
+        },
+      },
+    },
+    pi: {},
+    s: {
+      basePath,
+      originalBasePath: basePath,
+      canonicalProjectRoot: basePath,
+      resourceVersionOnStart: "test",
+      // Critical for the `!mid` canonical path: even when deriveState returns
+      // activeMilestone: null, the session still remembers the milestone it
+      // was working on, and we use that to look up DB status.
+      currentMilestoneId: "M001",
+      currentUnit: null,
+      milestoneMergedInPhases: false,
+      completionStopInProgress: overrides.completionStopInProgress,
+    },
+    prefs: undefined,
+    iteration: 1,
+    flowId: "test-flow",
+    nextSeq: () => 1,
+    deps: {
+      checkResourcesStale() {
+        return null;
+      },
+      invalidateAllCaches() {},
+      async preDispatchHealthGate() {
+        return { proceed: true, fixesApplied: [] };
+      },
+      async deriveState() {
+        return {
+          phase: "complete",
+          activeMilestone: overrides.activeMilestone,
+          activeSlice: null,
+          activeTask: null,
+          // Registry says M001 is complete and no other milestones exist, so
+          // `incomplete.length === 0 && state.registry.length > 0` evaluates
+          // true in the `!mid` branch.
+          registry: [{ id: "M001", status: "complete" }],
+          nextAction: "complete",
+        };
+      },
+      syncCmuxSidebar() {},
+      setActiveMilestoneId() {},
+      reconcileMergeState() {
+        return "clean";
+      },
+      // Anything below this point MUST NOT be reached when the guard fires.
+      preflightCleanRoot() {
+        recordSideEffect("preflight");
+        return { ok: true, stashPushed: false, stashMarker: null };
+      },
+      postflightPopStash() {
+        recordSideEffect("postflight");
+        return { ok: true, needsManualRecovery: false };
+      },
+      lifecycle: {
+        exitMilestone() {
+          recordSideEffect("merge");
+          return { ok: true };
+        },
+      },
+      sendDesktopNotification() {
+        recordSideEffect("desktop-notify");
+      },
+      logCmuxEvent() {
+        recordSideEffect("cmux-event");
+      },
+      async closeoutUnit() {
+        recordSideEffect("closeout-unit");
+      },
+      buildSnapshotOpts() {
+        return {};
+      },
+      async stopAuto(_ctx: unknown, _pi: unknown, reason?: string) {
+        recordSideEffect(`stop:${reason ?? ""}`);
+      },
+      async pauseAuto() {
+        recordSideEffect("pause");
+      },
+      emitJournalEvent() {
+        recordSideEffect("journal-event");
+      },
+    },
+  };
+}
+
+async function runScenario(opts: {
+  label: string;
+  completionStopInProgress: boolean;
+  activeMilestone: { id: string; title: string } | null;
+  // When true, the test opens an in-memory DB and inserts M001 with status
+  // "complete" so the DB-closed branch of the guard can fire.
+  dbAlreadyClosed: boolean;
+}): Promise<void> {
+  if (isDbAvailable()) {
+    closeDatabase();
+  }
+  if (opts.dbAlreadyClosed) {
+    openDatabase(":memory:");
+    insertMilestone({ id: "M001", title: "Milestone one", status: "complete" });
+  }
+
+  try {
+    const sideEffects: SideEffect[] = [];
+    const notifications: Array<{ message: string; level?: string }> = [];
+    const ic = makeIterationContext({
+      completionStopInProgress: opts.completionStopInProgress,
+      sideEffects,
+      notifications,
+      activeMilestone: opts.activeMilestone,
+    });
+
+    const result = await runPreDispatch(ic, {
+      recentUnits: [],
+      stuckRecoveryAttempts: 0,
+      consecutiveFinalizeTimeouts: 0,
+    });
+
+    assertTrue(
+      result.action === "break",
+      `${opts.label}: returns break instead of next`,
+    );
+    if (result.action === "break") {
+      assertTrue(
+        result.reason === "milestone-complete",
+        `${opts.label}: reason is milestone-complete (got "${result.reason}")`,
+      );
+    }
+    assertTrue(
+      sideEffects.length === 0,
+      `${opts.label}: no closeout side effects replayed (saw [${sideEffects.join(", ")}])`,
+    );
+    assertTrue(
+      notifications.length === 0,
+      `${opts.label}: no user notifications emitted (saw ${notifications.length})`,
+    );
+  } finally {
+    if (isDbAvailable()) {
+      closeDatabase();
+    }
+  }
+}
+
+console.log("\n=== Terminal complete is idempotent across both observer paths ===");
+
+// ── state.phase === "complete" branch (activeMilestone non-null) ────────────
+await runScenario({
+  label: "phase=complete + mid set + completionStopInProgress",
+  completionStopInProgress: true,
+  activeMilestone: { id: "M001", title: "Milestone one" },
+  dbAlreadyClosed: false,
+});
+
+await runScenario({
+  label: "phase=complete + mid set + DB closed",
+  completionStopInProgress: false,
+  activeMilestone: { id: "M001", title: "Milestone one" },
+  dbAlreadyClosed: true,
+});
+
+// ── Canonical !mid "all milestones complete" sub-branch ────────────────────
+// deriveState returns phase: "complete" with activeMilestone: null. The
+// session's s.currentMilestoneId (M001) is what the guard consults.
+await runScenario({
+  label: "phase=complete + activeMilestone=null + completionStopInProgress",
+  completionStopInProgress: true,
+  activeMilestone: null,
+  dbAlreadyClosed: false,
+});
+
+await runScenario({
+  label: "phase=complete + activeMilestone=null + DB closed",
+  completionStopInProgress: false,
+  activeMilestone: null,
+  dbAlreadyClosed: true,
+});
+
+report();

package/src/resources/extensions/gsd/tests/plan-gate-failed-doctor-heal-hint.test.ts

@@ -18,12 +18,12 @@ function readSrc(file: string): string {
   return readFileSync(join(gsdDir, file), "utf-8");
 }
 
-test("#4620: auto/phases plan gate failed message includes doctor heal hint", () => {
-  const src = readSrc("auto/phases.ts");
+test("#4620: auto/pre-dispatch plan gate failed message includes doctor heal hint", () => {
+  const src = readSrc("auto/pre-dispatch.ts");
   assert.match(
     src,
     /Plan gate failed-closed:[\s\S]*\/gsd doctor heal/,
-    "auto/phases.ts should include /gsd doctor heal in plan gate failed notification",
+    "auto/pre-dispatch.ts should include /gsd doctor heal in plan gate failed notification",
   );
 });