Keep /review and /ship working in copied Windows installs (GST-2347) (#4286)

* Keep /review and /ship working in copied Windows installs (GST-2347)

On Windows, setup copies bin/ and lib/ into each non-Claude host's runtime
folder, but left out the scripts/ modules they import. gstack-review-plan,
which /review and /ship run first, failed with "Cannot find module
'../scripts/resolvers/fix-check'". Setup now also copies fix-check.ts,
discover-skills.ts and lib/ci-history.ts, and the runtime-root test follows
every static import of the commands each host's skills run.

Co-authored-by: Satria Suria <57977752+SatriaSuriaAriffin@users.noreply.github.com>
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* Say the copied-install import check follows static imports only

The changelog entry claimed the new test follows every import; it follows static imports. TODOS.md now lists the three deferred commands with one remedy and records what the check does not see: lazy import() and require() calls, modules a shell script launches, and bare package imports.

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

* Keep the production-dependency check's walk sentinel on a module setup does not name

setup now names scripts/discover-skills.ts, so the check scans it as a root file and its import walk no longer reaches it. The walk's sentinel moves to scripts/gen-agents-digest.ts, which only gen-skill-docs imports, and discover-skills.ts stays asserted as scanned. Fixes the quick-bun-default failure in CI run 42857125482.

Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

---------

Co-authored-by: Satria Suria <57977752+SatriaSuriaAriffin@users.noreply.github.com>
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>

TODOS.md

@@ -425,10 +425,17 @@ From the approved 2026-10 test/eval/CI audit plan; the audit PR files these.
 - **`scripts/ship-measure.ts` on env-var hosts** — /ship's measure loop runs
   `$GSTACK_ROOT/scripts/ship-measure.ts`, which imports gstack's test helpers
   and eval harness, so a Codex/Factory/... runtime root would need most of the
-  development tree. It is the one `$GSTACK_ROOT` path
-  `test/runtime-root-assets.test.ts` defers. Either make ship-measure
-  self-contained or have the skill call it from the source checkout
-  (`.source-path`). **Effort:** M. **Priority:** P3.
+  development tree. `bin/gstack-model-benchmark` (/benchmark-models imports
+  the test tree's providers) and `bin/gstack-ship-queue` (the merge queue
+  imports the eval gate) have the same problem and fail to start from a
+  copied Windows root. `test/runtime-root-assets.test.ts` defers all three.
+  For each, either make it self-contained or have the skill call it from the
+  source checkout (`.source-path`). That test also follows static imports
+  only: lazy `import()` and `require()` calls (`bin/gstack-evidence` loads
+  `scripts/lib/shard-engine` with a fallback, and
+  `lib/pregate/flaky-exception.ts` requires it and a test helper), modules a
+  shell script launches, and bare package imports go unchecked.
+  **Effort:** M. **Priority:** P3.
 - **Rendered `SKILL.md` copies inside runtime roots are discoverable** —
   `office-hours/SKILL.md` (#2449) and now `plan-design-review/SKILL.md` sit in
   `~/.codex/skills/gstack/<skill>/` because skills read them as

changelog.d/keep-review-working-in-copied-windows-installs.md

@@ -0,0 +1,4 @@
+---
+important: blocker
+---
+- **/review and /ship start again on Windows under Codex, Cursor and the other non-Claude hosts.** On Windows, setup copies gstack's `bin/` and `lib/` into each host's runtime folder instead of linking them, and it left out three `scripts/` files those commands import. The review planner that /review and /ship run first stopped with `Cannot find module '../scripts/resolvers/fix-check'`, and the commands behind /retro and /write-skill had the same gap. Setup now copies `scripts/resolvers/fix-check.ts`, `scripts/discover-skills.ts` and `scripts/lib/ci-history.ts` too, and a test follows the static imports of the commands each host's skills run, so a new static import outside the runtime folder fails CI. Thanks to @SatriaSuriaAriffin, who found and reported the fix-check gap (#4275).

setup

@@ -790,9 +790,11 @@ _link_or_copy() {
 # residue): the preamble's jargon list, question registry and AskUserQuestion
 # docs, the review checklists and specialists, /plan-devex-review's Hall of
 # Fame, the test-value bar, design-html's vendored pretext.js and VERSION.
-# And crew/lib, which bin and lib import (the crew/ move, GST-1833): a root
-# whose bin and lib are copies (Windows) needs it beside them to start.
-# Pinned by test/runtime-root-assets.test.ts. A destination that already exists
+# And crew/lib, which bin and lib import (the crew/ move, GST-1833), plus the
+# scripts/ modules they import: a root whose bin and lib are copies (Windows)
+# needs them beside it to start (/review and /ship's review planner imports
+# fix-check.ts).
+# Pinned by test/runtime-root-assets.test.ts, which also follows those imports. A destination that already exists
 # as a real entry is left alone: in the repo-local sidecar review/ is a link to
 # the source, so its files already resolve and must never be linked over.
 _link_runtime_dists() {
@@ -813,6 +815,7 @@ _link_runtime_dists() {
     fi
   done
   for f in VERSION scripts/jargon-list.json scripts/question-registry.ts \
+    scripts/resolvers/fix-check.ts scripts/discover-skills.ts scripts/lib/ci-history.ts \
     docs/askuserquestion-split.md docs/askuserquestion-cjk.md docs/test-value-bar.md \
     review/checklist.md review/design-checklist.md review/greptile-triage.md review/TODOS-format.md \
     plan-devex-review/dx-hall-of-fame.md; do

test/runtime-root-assets.test.ts

@@ -20,6 +20,7 @@ import * as fs from 'fs';
 import * as os from 'os';
 import * as path from 'path';
 import { HOST_CONFIG_MAP } from '../hosts';
+import { bunIsolationArgs } from '../lib/bun-isolation';
 import { runGeneration } from '../scripts/gen-skill-docs';
 
 const ROOT = path.resolve(import.meta.dir, '..');
@@ -43,7 +44,9 @@ const ROOT_BUILDERS: Record<string, string> = {
 // Deferred, with the reason recorded in TODOS.md: /ship's measure loop runs
 // scripts/ship-measure.ts, which imports gstack's test helpers and eval
 // harness; a runtime root would need the whole development tree.
-const DEFERRED = new Set(['scripts/ship-measure.ts']);
+// gstack-model-benchmark (/benchmark-models) and the merge queue's
+// gstack-ship-queue are deferred the same way: both import that tree.
+const DEFERRED = new Set(['scripts/ship-measure.ts', 'bin/gstack-model-benchmark', 'bin/gstack-ship-queue']);
 
 // Compiled outputs: the stub source carries placeholder binaries, so a fresh
 // checkout without a build still exercises the links to them.
@@ -97,17 +100,52 @@ function makeSource(): void {
   }
 }
 
-function buildRoot(host: string): { root: string; stderr: string; status: number | null } {
-  const root = path.join(tmp, 'home', host, 'skills', 'gstack');
+// windows: setup's IS_WINDOWS=1 arm, where bin/, lib/ and the runtime dists
+// are real copies instead of links, so a module they import resolves only if
+// setup copied it beside them. It copies from the checkout itself, because
+// cp -R copies the stub source's top-level links as links.
+function buildRoot(host: string, windows = false): { root: string; stderr: string; status: number | null } {
+  const root = path.join(tmp, windows ? 'home-windows' : 'home', host, 'skills', 'gstack');
   fs.mkdirSync(path.dirname(root), { recursive: true });
   const functions = ['_link_or_copy', '_link_runtime_dists', '_copy_runtime_skill_refs', '_copy_skill_md',
     '_sidecar_root_user_owned', '_gstack_generated_header', ROOT_BUILDERS[host]!]
     .filter(name => SETUP_SRC.includes(`\n${name}() {`)).map(extractFunction).join('\n');
-  const r = spawnSync('bash', ['-c', `set -e\nIS_WINDOWS=0\n${functions}\n${ROOT_BUILDERS[host]} "${src}" "${root}"`],
+  const r = spawnSync('bash', ['-c', `set -e\nIS_WINDOWS=${windows ? 1 : 0}\n${functions}\n${ROOT_BUILDERS[host]} "${windows ? ROOT : src}" "${root}"`],
     { encoding: 'utf8', timeout: 30_000, env: { PATH: process.env.PATH ?? '/usr/bin:/bin', HOME: path.join(tmp, 'home') } });
   return { root, stderr: r.stderr, status: r.status };
 }
 
+const transpiler = new Bun.Transpiler({ loader: 'ts' });
+function isBunSource(file: string): boolean {
+  if (!fs.existsSync(file) || !fs.statSync(file).isFile()) return false;
+  return /\.(ts|js|mjs)$/.test(file) || (!path.extname(file) && /^#!.*\bbun\b/.test(fs.readFileSync(file, 'utf8')));
+}
+
+/**
+ * Static imports reachable from the code a host's skills run that do not
+ * resolve inside its runtime root. Paths resolve from where the root holds
+ * each file, never through its links, which is how a copied root loads them.
+ */
+function importsOutsideRoot(host: string, root: string): string[] {
+  const queue = [...referencedPaths(host)].filter(rel => !DEFERRED.has(rel)).map(rel => path.join(root, rel)).filter(isBunSource);
+  const seen = new Set<string>();
+  const outside = new Set<string>();
+  while (queue.length) {
+    const file = queue.pop()!;
+    if (seen.has(file)) continue;
+    seen.add(file);
+    for (const { path: spec, kind } of transpiler.scanImports(fs.readFileSync(file, 'utf8').replace(/^#!.*/, ''))) {
+      if (kind !== 'import-statement' || !spec.startsWith('.')) continue;
+      const target = path.resolve(path.dirname(file), spec);
+      const found = [target, target.replace(/\.js$/, '.ts'), `${target}.ts`, `${target}.js`, path.join(target, 'index.ts')]
+        .find(c => c.startsWith(root + path.sep) && fs.existsSync(c) && fs.statSync(c).isFile());
+      if (!found) outside.add(`${path.relative(root, file)} -> ${path.relative(root, target)}`);
+      else if (isBunSource(found)) queue.push(found);
+    }
+  }
+  return [...outside].sort();
+}
+
 beforeAll(async () => {
   const result = await runGeneration({ host: 'all', outputRoot: renderDir });
   if (result.exitCode !== 0) throw new Error(result.diagnostics.filter(d => d.kind === 'error').map(d => d.message).join('\n'));
@@ -139,6 +177,26 @@ describe.skipIf(process.platform === 'win32')('runtime roots carry every $GSTACK
     });
   }
 
+  for (const host of Object.keys(ROOT_BUILDERS)) {
+    test(`${host}: every module the skills' commands import is inside the runtime root`, () => {
+      const { root, status } = buildRoot(host);
+      expect(status).toBe(0);
+      expect(importsOutsideRoot(host, root)).toEqual([]);
+    });
+  }
+
+  test('the review planner starts from a copied (Windows) runtime root', () => {
+    const { root, stderr, status } = buildRoot('codex', true);
+    expect(stderr).toBe('');
+    expect(status).toBe(0);
+    expect(fs.lstatSync(path.join(root, 'bin')).isSymbolicLink()).toBe(false);
+    const r = spawnSync(process.execPath, [...bunIsolationArgs(), path.join(root, 'bin', 'gstack-review-plan'), '--help'],
+      { cwd: tmp, encoding: 'utf8', timeout: 30_000, env: { PATH: process.env.PATH ?? '/usr/bin:/bin', HOME: path.join(tmp, 'home-windows') } });
+    expect(r.stderr).not.toContain('Cannot find module');
+    expect(r.status).toBe(0);
+    expect(r.stdout).toContain('--fix-check-prompt');
+  });
+
   test('a runtime root never gains a whole skill directory beside the rendered copies', () => {
     const { root } = buildRoot('codex');
     for (const dir of ['plan-design-review', 'office-hours']) {

test/setup-production-deps.test.ts

@@ -572,7 +572,8 @@ describe('setup and runtime need only production dependencies', () => {
         queue.push(target);
       }
     }
-    expect(Object.keys(reached)).toContain('scripts/discover-skills.ts');
+    expect(Object.keys(reached)).toContain('scripts/gen-agents-digest.ts');
+    expect(Object.keys({ ...files, ...reached })).toContain('scripts/discover-skills.ts');
     expect(Object.keys(reached)).toContain('test/helpers/providers/claude.ts');
     expect(nonProductionImports(PKG, reached)).toEqual([]);
   });