@opengsd/gsd-pi 1.5.0 → 1.6.0

package/src/resources/extensions/gsd/tests/runtime-invariant-modules.test.ts

@@ -10,6 +10,7 @@ import {
   compileUnitContextContract,
   compileUnitToolContract,
   getUnitWorkflowDispatchReadinessError,
+  getUnitWorkflowDispatchReadinessErrorForModel,
 } from "../tool-contract.js";
 import { shouldBlockAutoUnitToolCall } from "../auto-unit-tool-scope.js";
 import type { GSDState } from "../types.js";
@@ -111,6 +112,25 @@ test("Tool Contract derives dispatch readiness from Unit workflow tools", () =>
   assert.equal(error, null);
 });
 
+test("Tool Contract derives dispatch readiness from selected model context", () => {
+  const authLookups: string[] = [];
+  const error = getUnitWorkflowDispatchReadinessErrorForModel({
+    model: { provider: "claude-code" },
+    fallbackModel: { provider: "openai-codex", baseUrl: "local://claude-code" },
+    getProviderAuthMode(provider) {
+      authLookups.push(provider);
+      return provider === "claude-code" ? "externalCli" : "oauth";
+    },
+    unitType: "plan-slice",
+    projectRoot: "/tmp/project",
+    env: { GSD_WORKFLOW_MCP_COMMAND: "node" },
+    surface: "auto-mode",
+  });
+
+  assert.equal(error, null);
+  assert.deepEqual(authLookups, ["claude-code"]);
+});
+
 test("Unit Context Contract exposes prompt context without workflow tool surface", () => {
   const result = compileUnitContextContract("execute-task");
 

package/src/resources/extensions/gsd/tests/sibling-model-fallbacks.test.ts

@@ -0,0 +1,189 @@
+/**
+ * Sibling single-model fields honor `fallbacks[]` (#1229).
+ *
+ * `post_unit_hooks[].model`, `auto_supervisor.model`, and
+ * `reactive_execution.subagent_model` accept the same
+ * `{ model, provider?, fallbacks? }` object form as `models.<phase>` so a
+ * transient provider trip falls back to the next entry instead of hard-failing
+ * (and, for a blocking hook, pausing the whole auto-mode run).
+ */
+
+import test from "node:test";
+import assert from "node:assert/strict";
+import { mkdtempSync, rmSync, writeFileSync } from "node:fs";
+import { join } from "node:path";
+import { tmpdir } from "node:os";
+
+import {
+  normalizeModelFieldConfig,
+  resolveModelWithFallbacksForUnit,
+  resolveAutoSupervisorConfig,
+} from "../preferences-models.ts";
+import { validatePreferences } from "../preferences-validation.ts";
+
+function withPreferences<T>(frontmatter: string[], fn: () => T): T {
+  const oldHome = process.env.GSD_HOME;
+  const home = mkdtempSync(join(tmpdir(), "gsd-sibling-models-"));
+  try {
+    process.env.GSD_HOME = home;
+    writeFileSync(join(home, "preferences.md"), ["---", ...frontmatter, "---", ""].join("\n"));
+    return fn();
+  } finally {
+    if (oldHome === undefined) delete process.env.GSD_HOME;
+    else process.env.GSD_HOME = oldHome;
+    rmSync(home, { recursive: true, force: true });
+  }
+}
+
+// ─── normalizeModelFieldConfig ───────────────────────────────────────────────
+
+test("normalizeModelFieldConfig: bare string → primary with empty fallbacks", () => {
+  assert.deepEqual(normalizeModelFieldConfig("my-model"), { primary: "my-model", fallbacks: [] });
+});
+
+test("normalizeModelFieldConfig: object form carries fallbacks", () => {
+  assert.deepEqual(
+    normalizeModelFieldConfig({ model: "primary", fallbacks: ["fb-1", "fb-2"] }),
+    { primary: "primary", fallbacks: ["fb-1", "fb-2"] },
+  );
+});
+
+test("normalizeModelFieldConfig: provider prefix is prepended to bare model", () => {
+  assert.deepEqual(
+    normalizeModelFieldConfig({ model: "gpt-5.4", provider: "openai-codex" }),
+    { primary: "openai-codex/gpt-5.4", fallbacks: [] },
+  );
+});
+
+test("normalizeModelFieldConfig: unset / blank / model-less → undefined", () => {
+  assert.equal(normalizeModelFieldConfig(undefined), undefined);
+  assert.equal(normalizeModelFieldConfig("   "), undefined);
+  assert.equal(normalizeModelFieldConfig({ model: "" }), undefined);
+});
+
+// ─── post_unit_hooks[].model fallback resolution ─────────────────────────────
+
+test("hook/<name> resolves the hook's object-form model with fallbacks", () => {
+  withPreferences(
+    [
+      "post_unit_hooks:",
+      "  - name: plan-review",
+      "    after: [plan-slice]",
+      "    prompt: review the plan",
+      "    model:",
+      "      model: primary-model",
+      "      fallbacks:",
+      "        - fallback-a",
+      "        - fallback-b",
+    ],
+    () => {
+      const resolved = resolveModelWithFallbacksForUnit("hook/plan-review");
+      assert.deepEqual(resolved, { primary: "primary-model", fallbacks: ["fallback-a", "fallback-b"] });
+    },
+  );
+});
+
+test("hook/<name> still resolves the legacy bare-string model form", () => {
+  withPreferences(
+    [
+      "post_unit_hooks:",
+      "  - name: plan-review",
+      "    after: [plan-slice]",
+      "    prompt: review the plan",
+      "    model: solo-model",
+    ],
+    () => {
+      assert.deepEqual(
+        resolveModelWithFallbacksForUnit("hook/plan-review"),
+        { primary: "solo-model", fallbacks: [] },
+      );
+    },
+  );
+});
+
+test("hook/<name> resolves undefined when the hook has no model", () => {
+  withPreferences(
+    [
+      "post_unit_hooks:",
+      "  - name: plan-review",
+      "    after: [plan-slice]",
+      "    prompt: review the plan",
+    ],
+    () => {
+      assert.equal(resolveModelWithFallbacksForUnit("hook/plan-review"), undefined);
+      assert.equal(resolveModelWithFallbacksForUnit("hook/unknown-hook"), undefined);
+    },
+  );
+});
+
+// ─── auto_supervisor.model ───────────────────────────────────────────────────
+
+test("auto_supervisor.model object form resolves with fallbacks", () => {
+  withPreferences(
+    [
+      "auto_supervisor:",
+      "  model:",
+      "    model: supervisor-primary",
+      "    fallbacks: [supervisor-fb]",
+    ],
+    () => {
+      assert.deepEqual(resolveModelWithFallbacksForUnit("supervisor"), {
+        primary: "supervisor-primary",
+        fallbacks: ["supervisor-fb"],
+      });
+      const supervisor = resolveAutoSupervisorConfig();
+      assert.equal(supervisor.model, "supervisor-primary");
+      assert.deepEqual(supervisor.modelFallbacks, ["supervisor-fb"]);
+    },
+  );
+});
+
+// ─── validation accepts the object form ──────────────────────────────────────
+
+test("validatePreferences accepts object-form hook and subagent models", () => {
+  const { preferences: validated, errors } = validatePreferences({
+    post_unit_hooks: [
+      {
+        name: "plan-review",
+        after: ["plan-slice"],
+        prompt: "review",
+        model: { model: "primary", fallbacks: ["fb-1"] },
+      },
+    ],
+    reactive_execution: {
+      enabled: true,
+      subagent_model: { model: "sub-primary", fallbacks: ["sub-fb"] },
+    },
+  } as never);
+
+  assert.deepEqual(errors, []);
+  assert.deepEqual(validated.post_unit_hooks?.[0].model, { model: "primary", fallbacks: ["fb-1"] });
+  assert.deepEqual(validated.reactive_execution?.subagent_model, { model: "sub-primary", fallbacks: ["sub-fb"] });
+});
+
+test("validatePreferences rejects a model-less object form", () => {
+  const { errors } = validatePreferences({
+    post_unit_hooks: [
+      { name: "plan-review", after: ["plan-slice"], prompt: "review", model: { fallbacks: ["fb"] } },
+    ],
+  } as never);
+  assert.ok(errors.some((e) => e.includes('requires a non-empty "model"')));
+});
+
+test("validatePreferences accepts object-form auto_supervisor.model", () => {
+  const { preferences: validated, errors } = validatePreferences({
+    auto_supervisor: { model: { model: "sup-primary", fallbacks: ["sup-fb"] } },
+  } as never);
+  assert.deepEqual(errors, []);
+  assert.deepEqual(validated.auto_supervisor?.model, { model: "sup-primary", fallbacks: ["sup-fb"] });
+});
+
+test("validatePreferences rejects a model-less auto_supervisor.model object", () => {
+  const { errors } = validatePreferences({
+    auto_supervisor: { model: { fallbacks: ["sup-fb"] } },
+  } as never);
+  assert.ok(
+    errors.some((e) => e.includes("auto_supervisor.model") && e.includes('requires a non-empty "model"')),
+    `expected supervisor model error, got: ${JSON.stringify(errors)}`,
+  );
+});

package/src/resources/extensions/gsd/tests/single-field-model-thinking.test.ts

@@ -0,0 +1,156 @@
+/**
+ * Regression tests for #1269 — the object model form's `thinking` sub-field on
+ * the three single-model fields (`post_unit_hooks[].model`,
+ * `auto_supervisor.model`, and `reactive_execution.subagent_model`) was
+ * silently dropped after validation/normalization, so a configured
+ * hook/supervisor/subagent reasoning level never applied.
+ *
+ * These exercise the real preference pipeline through the exported runtime APIs:
+ *   sanitize   → validatePreferences
+ *   normalize  → normalizeModelFieldConfig
+ *   resolve    → resolveThinkingLevelForUnit  (supervisor + hook/* early path)
+ *
+ * The `supervisor` and `hook/*` unit types have no phase-bucket chain, so the
+ * pre-#1269 resolver walk never covered them and returned `undefined`.
+ */
+
+import test 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 { validatePreferences } from "../preferences-validation.ts";
+import { normalizeModelFieldConfig, resolveThinkingLevelForUnit } from "../preferences-models.ts";
+import { invalidateAllCaches } from "../cache.ts";
+
+/** Read the per-field `thinking` off a validated/normalized model config. */
+function thinkingOf(model: unknown): string | undefined {
+  return (model as { thinking?: string } | undefined)?.thinking;
+}
+
+/** Write a project-level `.gsd/PREFERENCES.md` and run `fn` with its basePath. */
+function withProjectPreferences<T>(frontmatter: string, fn: (basePath: string) => T): T {
+  const base = mkdtempSync(join(tmpdir(), "gsd-1269-thinking-"));
+  mkdirSync(join(base, ".gsd"), { recursive: true });
+  writeFileSync(join(base, ".gsd", "PREFERENCES.md"), `---\n${frontmatter}\n---\n`, "utf-8");
+  invalidateAllCaches();
+  try {
+    return fn(base);
+  } finally {
+    invalidateAllCaches();
+    rmSync(base, { recursive: true, force: true });
+  }
+}
+
+// ── sanitize: validatePreferences preserves a valid object-form thinking ──────
+
+test("validatePreferences preserves auto_supervisor.model object-form thinking", () => {
+  const { preferences, errors } = validatePreferences({
+    auto_supervisor: { model: { model: "supervisor-model", thinking: "high" } },
+  });
+  assert.equal(errors.length, 0);
+  assert.equal(thinkingOf(preferences.auto_supervisor?.model), "high");
+});
+
+test("validatePreferences preserves post_unit_hooks[].model object-form thinking", () => {
+  const { preferences, errors } = validatePreferences({
+    post_unit_hooks: [
+      {
+        name: "code-review",
+        after: ["execute-task"],
+        prompt: "Review the diff",
+        model: { model: "hook-model", thinking: "low" },
+      },
+    ],
+  });
+  assert.equal(errors.length, 0);
+  assert.equal(thinkingOf(preferences.post_unit_hooks?.[0]?.model), "low");
+});
+
+test("validatePreferences preserves reactive_execution.subagent_model object-form thinking", () => {
+  const { preferences, errors } = validatePreferences({
+    reactive_execution: {
+      enabled: true,
+      max_parallel: 2,
+      isolation_mode: "same-tree",
+      subagent_model: { model: "subagent-model", thinking: "medium" },
+    },
+  });
+  assert.equal(errors.length, 0);
+  assert.equal(thinkingOf(preferences.reactive_execution?.subagent_model), "medium");
+});
+
+// ── sanitize: an invalid thinking level is warned and stripped ────────────────
+
+test("validatePreferences warns and strips an invalid object-form thinking level", () => {
+  const { preferences, errors, warnings } = validatePreferences({
+    auto_supervisor: { model: { model: "supervisor-model", thinking: "ludicrous" } },
+  } as never);
+  // The model survives; only the bogus thinking is dropped, so a typo can never
+  // reach the resolver and masquerade as explicit configuration.
+  assert.equal(errors.length, 0);
+  assert.equal(thinkingOf(preferences.auto_supervisor?.model), undefined);
+  assert.ok(
+    warnings.some((w) => w.includes("auto_supervisor.model") && w.includes("thinking")),
+    `expected a thinking warning, got: ${JSON.stringify(warnings)}`,
+  );
+});
+
+// ── normalize: normalizeModelFieldConfig carries thinking through ─────────────
+
+test("normalizeModelFieldConfig carries object-form thinking through", () => {
+  const config = normalizeModelFieldConfig({ model: "hook-model", thinking: "xhigh" });
+  assert.equal(config?.primary, "hook-model");
+  assert.equal(config?.thinking, "xhigh");
+});
+
+test("normalizeModelFieldConfig leaves thinking absent when omitted", () => {
+  const fromObject = normalizeModelFieldConfig({ model: "hook-model" });
+  assert.equal(fromObject?.primary, "hook-model");
+  assert.equal(fromObject?.thinking, undefined);
+
+  const fromString = normalizeModelFieldConfig("hook-model");
+  assert.equal(fromString?.primary, "hook-model");
+  assert.equal(fromString?.thinking, undefined);
+});
+
+// ── resolve: the single-field unit types surface their per-field thinking ──────
+
+test("resolveThinkingLevelForUnit('supervisor') returns auto_supervisor.model thinking", () => {
+  withProjectPreferences(
+    ["auto_supervisor:", "  model:", "    model: supervisor-model", "    thinking: high"].join("\n"),
+    (base) => {
+      assert.equal(resolveThinkingLevelForUnit("supervisor", base), "high");
+    },
+  );
+});
+
+test("resolveThinkingLevelForUnit('hook/<name>') returns the matching hook model thinking", () => {
+  withProjectPreferences(
+    [
+      "post_unit_hooks:",
+      "  - name: code-review",
+      "    after:",
+      "      - execute-task",
+      "    prompt: Review the diff",
+      "    model:",
+      "      model: hook-model",
+      "      thinking: low",
+    ].join("\n"),
+    (base) => {
+      assert.equal(resolveThinkingLevelForUnit("hook/code-review", base), "low");
+    },
+  );
+});
+
+test("resolveThinkingLevelForUnit('supervisor') stays undefined when thinking is omitted", () => {
+  // Bare-string model form carries no thinking → dispatch falls back to the
+  // session/floor level, unchanged from pre-#1269 behavior.
+  withProjectPreferences(
+    ["auto_supervisor:", "  model: supervisor-model"].join("\n"),
+    (base) => {
+      assert.equal(resolveThinkingLevelForUnit("supervisor", base), undefined);
+    },
+  );
+});

package/src/resources/extensions/gsd/tests/skill-discovery.test.ts

@@ -109,3 +109,49 @@ test("appendDiscoveredSkillsFallback does not duplicate skills already in the pr
     location: "/tmp/already-loaded/SKILL.md",
   }]), prompt);
 });
+
+test("snapshotSkills({cwd}) detects skills added to the project .agents/skills/ mid-session", async () => {
+  await withTempSkillHome(async (home) => {
+    // Use a project dir under the temp home so cwd is isolated.
+    const projectDir = join(home, "myproject");
+    const projectSkills = join(projectDir, ".agents", "skills");
+    snapshotSkills({ cwd: projectDir });
+
+    // Add a skill to the project skills dir after the snapshot.
+    writeDiskSkill(projectSkills, "project-disk-skill", "Project skill added mid-session.");
+    const detected = detectNewSkills();
+
+    assert.deepEqual(detected.map(skill => skill.name), ["project-disk-skill"]);
+    assert.equal(detected[0].location, join(projectSkills, "project-disk-skill", "SKILL.md"));
+  });
+});
+
+test("snapshotSkills({cwd}) detects skills added to the project .claude/skills/ mid-session", async () => {
+  await withTempSkillHome(async (home) => {
+    const projectDir = join(home, "claude-project");
+    const claudeSkills = join(projectDir, ".claude", "skills");
+    snapshotSkills({ cwd: projectDir });
+
+    writeDiskSkill(claudeSkills, "claude-project-skill", "Claude-format skill added mid-session.");
+    const detected = detectNewSkills();
+
+    assert.deepEqual(detected.map(skill => skill.name), ["claude-project-skill"]);
+  });
+});
+
+test("snapshotSkills({cwd}) treats skills already in the project dir as baseline, not newly discovered", async () => {
+  await withTempSkillHome(async (home) => {
+    // projectDir is intentionally NOT process.cwd(): mirrors auto-mode passing a
+    // base path that differs from cwd (e.g. after chdir into a worktree). The
+    // baseline must scan the passed cwd's dirs, so a skill present *before* the
+    // snapshot is part of the baseline and is never reported as newly discovered.
+    // Regresses the fix where snapshotCurrentSkillNames fell back to process.cwd().
+    const projectDir = join(home, "preexisting-project");
+    const projectSkills = join(projectDir, ".agents", "skills");
+    writeDiskSkill(projectSkills, "preexisting-project-skill");
+
+    snapshotSkills({ cwd: projectDir });
+
+    assert.deepEqual(detectNewSkills(), []);
+  });
+});

package/src/resources/extensions/gsd/tests/skipped-validation-db-atomicity.test.ts

@@ -52,7 +52,12 @@ test("skipped validation dispatch persists the validation file and DB assessment
     const artifactPath = "milestones/M001/slices/S01/S01-ASSESSMENT.md";
     const assessmentPath = `.gsd/${artifactPath}`;
     assert.equal(getArtifact(artifactPath)?.artifact_type, "ASSESSMENT");
-    assert.equal(getAssessment(assessmentPath)?.scope, "run-uat");
+    // #1258: the per-slice ASSESSMENT fabricated by the pre-validation backfill
+    // is a placeholder (no UAT ever ran for this slice), so it is filed under the
+    // `backfill` scope — never `run-uat` — so `readUatGateVerdict` cannot mistake
+    // it for a genuine UAT sign-off. Its verdict body stays PASS to satisfy the
+    // MV02 artifact-existence requirement.
+    assert.equal(getAssessment(assessmentPath)?.scope, "backfill");
     assert.equal(getAssessment(assessmentPath)?.status, "pass");
     assert.equal(
       getLatestAssessmentByScope("M001", "milestone-validation")?.status,

package/src/resources/extensions/gsd/tests/stale-worktree-cwd.test.ts

@@ -12,11 +12,9 @@ import { join, sep } from "node:path";
 import { tmpdir } from "node:os";
 import { execSync } from "node:child_process";
 
-import {
-  createAutoWorktree,
-  teardownAutoWorktree,
-  mergeMilestoneToMain,
-} from "../auto-worktree.ts";
+import { createAutoWorktree } from "../auto-worktree-creation.ts";
+import { mergeMilestoneToMain } from "../auto-worktree-merge.ts";
+import { teardownAutoWorktree } from "../auto-worktree-teardown.ts";
 import { _resetServiceCache } from "../worktree.ts";
 import { _clearGsdRootCache } from "../paths.ts";
 

package/src/resources/extensions/gsd/tests/start-auto-detached.test.ts

@@ -208,6 +208,38 @@ test("startAutoDetached reports failures asynchronously (#3733)", () => {
   );
 });
 
+test("failed auto-start always clears leaked s.active so later /gsd auto is not bricked (#1235)", () => {
+  const autoSrc = readGsdFile("auto.ts");
+
+  // 1. startAutoDetached is the single funnel for any exception thrown out of
+  //    startAuto — including a bootstrap throw that escapes before the loop's
+  //    try/finally could run cleanupAfterLoopExit. The failure handler must
+  //    clear leaked activation, or s.active + the on-disk auto lock persist and
+  //    the re-entry guard silently no-ops every later /gsd auto until restart.
+  const detachedIdx = autoSrc.indexOf("export function startAutoDetached");
+  assert.ok(detachedIdx > -1, "startAutoDetached should exist");
+  const detachedBody = autoSrc.slice(detachedIdx, autoSrc.indexOf("\n}", detachedIdx));
+  assert.ok(
+    detachedBody.includes("if (s.active)") &&
+      detachedBody.includes("cleanupAfterLoopExit(ctx)"),
+    "startAutoDetached failure handler must run cleanupAfterLoopExit when auto-mode is still active",
+  );
+
+  // 2. bootstrapAutoSession sets s.active = true before some bail-outs return
+  //    false (e.g. the SQLite-unavailable gate), so the fresh-start !ready path
+  //    must clear the leaked activation before returning.
+  const startAutoIdx = autoSrc.indexOf("export async function startAuto(");
+  assert.ok(startAutoIdx > -1, "startAuto should exist");
+  const startAutoBody = autoSrc.slice(startAutoIdx);
+  const notReadyIdx = startAutoBody.indexOf("if (!ready) {");
+  assert.ok(notReadyIdx > -1, "fresh start should guard on bootstrap readiness with a block body");
+  const notReadyBlock = startAutoBody.slice(notReadyIdx, notReadyIdx + 600);
+  assert.ok(
+    notReadyBlock.includes("if (s.active) await cleanupAfterLoopExit(ctx)"),
+    "fresh-start !ready bail-out must clear leaked auto activation before returning",
+  );
+});
+
 test("detached auto-start keeps a ref'ed handle until the run settles", async () => {
   const originalSetInterval = globalThis.setInterval;
   const originalClearInterval = globalThis.clearInterval;

package/src/resources/extensions/gsd/tests/stash-pop-gsd-conflict.test.ts

@@ -14,7 +14,8 @@ import { join } from "node:path";
 import { tmpdir } from "node:os";
 import { execSync } from "node:child_process";
 
-import { createAutoWorktree, mergeMilestoneToMain } from "../auto-worktree.ts";
+import { createAutoWorktree } from "../auto-worktree-creation.ts";
+import { mergeMilestoneToMain } from "../auto-worktree-merge.ts";
 import { _resetServiceCache } from "../worktree.ts";
 import { _clearGsdRootCache } from "../paths.ts";
 

package/src/resources/extensions/gsd/tests/stash-queued-context-files.test.ts

@@ -28,7 +28,8 @@ import { join } from "node:path";
 import { tmpdir } from "node:os";
 import { execSync } from "node:child_process";
 
-import { createAutoWorktree, mergeMilestoneToMain } from "../auto-worktree.ts";
+import { createAutoWorktree } from "../auto-worktree-creation.ts";
+import { mergeMilestoneToMain } from "../auto-worktree-merge.ts";
 import { _resetServiceCache } from "../worktree.ts";
 import { _clearGsdRootCache } from "../paths.ts";
 

package/src/resources/extensions/gsd/tests/steer-worktree-path.test.ts

@@ -8,7 +8,7 @@ import { mkdtempSync, mkdirSync, existsSync, readFileSync, rmSync, writeFileSync
 import { join } from "node:path";
 import { tmpdir } from "node:os";
 import { appendOverride, loadActiveOverrides } from "../files.ts";
-import { getAutoWorktreePath } from "../auto-worktree.ts";
+import { getAutoWorktreePath } from "../auto-worktree-path-resolution.ts";
 
 describe("steer worktree path resolution (#3476)", () => {
   let projectRoot: string;