diff --git a/browse/test/gstack-config.test.ts b/browse/test/gstack-config.test.ts
index e342a50b7..670c1b092 100644
--- a/browse/test/gstack-config.test.ts
+++ b/browse/test/gstack-config.test.ts
@@ -15,12 +15,20 @@ const SCRIPT = join(import.meta.dir, '..', '..', 'bin', 'gstack-config');
 let stateDir: string;
 
 function run(args: string[] = [], extraEnv: Record<string, string> = {}) {
+  // The script resolves GSTACK_STATE_ROOT > GSTACK_HOME > GSTACK_STATE_DIR.
+  // Scrub the two higher-priority overrides from the inherited env so a
+  // leaked GSTACK_HOME/GSTACK_STATE_ROOT from another test file (or the
+  // operator shell) can never redirect reads/writes away from this test's
+  // temp stateDir.
+  const env: Record<string, string | undefined> = {
+    ...process.env,
+    GSTACK_STATE_DIR: stateDir,
+    ...extraEnv,
+  };
+  if (!('GSTACK_STATE_ROOT' in extraEnv)) delete env.GSTACK_STATE_ROOT;
+  if (!('GSTACK_HOME' in extraEnv)) delete env.GSTACK_HOME;
   const result = Bun.spawnSync(['bash', SCRIPT, ...args], {
-    env: {
-      ...process.env,
-      GSTACK_STATE_DIR: stateDir,
-      ...extraEnv,
-    },
+    env,
     stdout: 'pipe',
     stderr: 'pipe',
   });
diff --git a/test/codex-resume-flag-semantics.test.ts b/test/codex-resume-flag-semantics.test.ts
index 9075e0489..da866418b 100644
--- a/test/codex-resume-flag-semantics.test.ts
+++ b/test/codex-resume-flag-semantics.test.ts
@@ -15,8 +15,21 @@
 import { describe, test, expect } from 'bun:test';
 import { spawnSync } from 'child_process';
 
+// "On PATH" is not enough: a broken install (e.g. the npm wrapper missing its
+// platform binary @openai/codex-darwin-arm64) resolves via `which` but can't
+// answer any subcommand, so every assertion would fail on wrapper stack traces
+// instead of real flag-semantics regressions. Probe `codex --version` and only
+// run when the CLI actually executes; a working codex still gets the full
+// assertions.
 const codexPath = spawnSync('which', ['codex'], { encoding: 'utf-8' }).stdout.trim();
-const codexAvailable = codexPath.length > 0;
+const codexRuns =
+  codexPath.length > 0 &&
+  spawnSync('codex', ['--version'], {
+    encoding: 'utf-8',
+    stdio: ['ignore', 'pipe', 'pipe'],
+    timeout: 10_000,
+  }).status === 0;
+const codexAvailable = codexRuns;
 
 describe.skipIf(!codexAvailable)(
   'codex exec resume — flag semantics (live CLI smoke; closes #1270 regex-only gap)',
diff --git a/test/gbrain-detect-install.test.ts b/test/gbrain-detect-install.test.ts
index b9c82c155..fea588916 100644
--- a/test/gbrain-detect-install.test.ts
+++ b/test/gbrain-detect-install.test.ts
@@ -25,7 +25,15 @@ const INSTALL = path.join(ROOT, 'bin', 'gstack-gbrain-install');
 // dirs — this keeps `gbrain` out of PATH deterministically across dev machines
 // while still finding jq, git, curl, sed, cat, etc. Each test can prepend a
 // fake-gbrain dir when it wants to simulate presence.
-const SAFE_PATH = '/usr/bin:/bin:/usr/sbin:/sbin:/opt/homebrew/bin:/usr/local/bin';
+//
+// One wrinkle: bin/gstack-gbrain-detect's shebang is `#!/usr/bin/env -S bun run`,
+// so the child's PATH must contain `bun`. On dev machines bun usually lives in
+// ~/.bun/bin, which we deliberately exclude (it also holds a real `gbrain` from
+// `bun link`). Expose ONLY the running bun interpreter through a shim dir,
+// appended last so system dirs still win for everything else.
+const BUN_SHIM_DIR = fs.mkdtempSync(path.join(os.tmpdir(), 'bun-shim-'));
+fs.symlinkSync(process.execPath, path.join(BUN_SHIM_DIR, 'bun'));
+const SAFE_PATH = `/usr/bin:/bin:/usr/sbin:/sbin:/opt/homebrew/bin:/usr/local/bin:${BUN_SHIM_DIR}`;
 
 let tmpHome: string;
 let tmpHomeReal: string;
diff --git a/test/gen-skill-docs.test.ts b/test/gen-skill-docs.test.ts
index 2fb783ffd..8d23b4be0 100644
--- a/test/gen-skill-docs.test.ts
+++ b/test/gen-skill-docs.test.ts
@@ -177,6 +177,13 @@ describe('gen-skill-docs', () => {
   // whose plain `description:` scalar contains an interior ": " (read as a nested
   // mapping). Parse EVERY generated frontmatter block with a strict YAML parser,
   // not just string-check that name:/description: exist.
+  //
+  // Bun.YAML landed in bun 1.3; older local bun has no in-process strict YAML
+  // parser (the repo is deliberately zero-dep for YAML). CI pins
+  // `bun-version: latest`, so coverage is guaranteed there — on an old local
+  // bun we skip rather than false-fail.
+  const hasBunYaml = typeof (Bun as any).YAML?.parse === 'function';
+
   function frontmatterBlock(content: string): string {
     expect(content.startsWith('---\n')).toBe(true);
     const end = content.indexOf('\n---', 4);
@@ -184,7 +191,7 @@ describe('gen-skill-docs', () => {
     return content.slice(4, end);
   }
 
-  test('every generated SKILL.md frontmatter parses as strict YAML', () => {
+  test.skipIf(!hasBunYaml)('every generated SKILL.md frontmatter parses as strict YAML', () => {
     for (const skill of CLAUDE_GENERATED_SKILLS) {
       const content = fs.readFileSync(path.join(ROOT, skill.dir, 'SKILL.md'), 'utf-8');
       const fm = frontmatterBlock(content);
@@ -196,7 +203,7 @@ describe('gen-skill-docs', () => {
     }
   });
 
-  test('every generated Codex (.agents/skills) frontmatter parses as strict YAML', () => {
+  test.skipIf(!hasBunYaml)('every generated Codex (.agents/skills) frontmatter parses as strict YAML', () => {
     const agentsDir = path.join(ROOT, '.agents', 'skills');
     if (!fs.existsSync(agentsDir)) return; // skip if external hosts not generated
     for (const entry of fs.readdirSync(agentsDir, { withFileTypes: true })) {
diff --git a/test/gstack-memory-helpers.test.ts b/test/gstack-memory-helpers.test.ts
index 7b356a99c..9f190732f 100644
--- a/test/gstack-memory-helpers.test.ts
+++ b/test/gstack-memory-helpers.test.ts
@@ -249,11 +249,14 @@ body
 // ── withErrorContext ───────────────────────────────────────────────────────
 
 describe("withErrorContext", () => {
-  let savedHome: string | undefined;
+  // Captured ONCE at collection time, NOT in beforeEach. Saving inside
+  // beforeEach clobbers the real value with the previous test's temp dir
+  // after test #1, so the afterAll restore leaked a dead temp path into
+  // process.env for every test file that ran after this one.
+  const savedHome = process.env.GSTACK_HOME;
   let testHome: string;
 
   beforeEach(() => {
-    savedHome = process.env.GSTACK_HOME;
     testHome = mkdtempSync(join(tmpdir(), "gstack-test-home-"));
     process.env.GSTACK_HOME = testHome;
   });
@@ -312,18 +315,20 @@ describe("withErrorContext", () => {
 // ── detectEngineTier ───────────────────────────────────────────────────────
 
 describe("detectEngineTier", () => {
-  let savedHome: string | undefined;
-  let savedGbrainHome: string | undefined;
-  let savedRealHome: string | undefined;
-  let savedPath: string | undefined;
+  // Captured ONCE at collection time, NOT in beforeEach. Saving inside
+  // beforeEach clobbers the real values with the previous test's temp dirs
+  // after test #1, so the afterAll restore leaked dead gstack-test-engine-*
+  // temp paths into GSTACK_HOME / GBRAIN_HOME / HOME for every test file
+  // that ran after this one (broke gstack-config + make-pdf format-gate in
+  // full-suite runs).
+  const savedHome = process.env.GSTACK_HOME;
+  const savedGbrainHome = process.env.GBRAIN_HOME;
+  const savedRealHome = process.env.HOME;
+  const savedPath = process.env.PATH;
   let testHome: string;
   let testGbrainHome: string;
 
   beforeEach(() => {
-    savedHome = process.env.GSTACK_HOME;
-    savedGbrainHome = process.env.GBRAIN_HOME;
-    savedRealHome = process.env.HOME;
-    savedPath = process.env.PATH;
     testHome = mkdtempSync(join(tmpdir(), "gstack-test-engine-"));
     testGbrainHome = mkdtempSync(join(tmpdir(), "gstack-test-gbrain-"));
     process.env.GSTACK_HOME = testHome;
diff --git a/test/openclaw-native-skills.test.ts b/test/openclaw-native-skills.test.ts
index 009b5e22c..539a677e7 100644
--- a/test/openclaw-native-skills.test.ts
+++ b/test/openclaw-native-skills.test.ts
@@ -18,8 +18,14 @@ function extractFrontmatter(content: string): string {
   return content.slice(4, fmEnd);
 }
 
+// Bun.YAML landed in bun 1.3; older local bun has no in-process strict YAML
+// parser (the repo is deliberately zero-dep for YAML). CI pins
+// `bun-version: latest`, so coverage is guaranteed there — on an old local
+// bun we skip rather than false-fail.
+const hasBunYaml = typeof (Bun as any).YAML?.parse === 'function';
+
 describe('OpenClaw native skills', () => {
-  test('frontmatter parses as YAML and keeps only name + description', () => {
+  test.skipIf(!hasBunYaml)('frontmatter parses as YAML and keeps only name + description', () => {
     for (const skill of OPENCLAW_NATIVE_SKILLS) {
       const content = fs.readFileSync(path.join(ROOT, skill), 'utf-8');
       const frontmatter = extractFrontmatter(content);