fix(benchmark): skip-trust for headless gemini + drop release metadata

Pass --skip-trust so temp/untrusted workdirs work in headless
benchmarks (mirrors Codex --skip-git-repo-check). Revert VERSION,
CHANGELOG, and package.json bumps — leave release bookkeeping to
maintainers / /ship, matching community PR practice.

Co-authored-by: Cursor <cursoragent@cursor.com>

CHANGELOG.md

@@ -1,23 +1,5 @@
 # Changelog
 
-## [1.60.2.0] - 2026-07-10
-
-## **Gemini model benchmarks no longer rank a silent empty run as a $0 win.**
-
-`gstack-model-benchmark --models gemini` was parsing an older stream-json shape (`message.text`, `result.usage`). Current gemini CLI emits `message.content` (with `role`) and `result.stats`, so the adapter returned empty output, 0 tokens, and no error — a fake success that quietly mis-ranked gemini in comparison tables.
-
-### What this means for you
-
-Gemini benchmark rows now show real assistant text and token counts again. If a run somehow still exits 0 with empty output, it surfaces as an error row instead of a $0 success.
-
-### Itemized changes
-
-#### Fixed
-
-- `test/helpers/providers/gemini.ts`: `parseGeminiStreamJson` accepts current CLI `content` + `role:'assistant'` (ignores user-prompt echoes), reads tokens from `result.stats` with legacy `usage` fallbacks, and takes `model` from `init`. Empty exit-0 output is reported as `error.code=unknown` instead of a silent success (#2159).
-- `test/helpers/providers/gemini.test.ts`: unit coverage for current + legacy stream shapes, user-echo guard, and empty parse.
-- `test/skill-e2e-benchmark-providers.test.ts`: live gemini smoke asserts non-empty output containing the probe string and positive token counts.
-
 ## [1.60.1.0] - 2026-07-09
 
 ## **The /autoplan dual-voice eval is back on the board, catching real regressions.**

VERSION

@@ -1 +1 @@
-1.60.2.0
+1.60.1.0

package.json

@@ -1,6 +1,6 @@
 {
   "name": "gstack",
-  "version": "1.60.2.0",
+  "version": "1.60.1.0",
   "description": "Garry's Stack — Claude Code skills + fast headless browser. One repo, one install, entire AI engineering workflow.",
   "license": "MIT",
   "type": "module",

test/helpers/gemini-session-runner.ts

@@ -8,7 +8,8 @@
  * Key differences from Codex session-runner:
  * - Uses `gemini -p` instead of `codex exec`
  * - Output is NDJSON with event types: init, message, tool_use, tool_result, result
- * - Uses `--output-format stream-json --yolo` instead of `--json -s read-only`
+ * - Uses `--output-format stream-json --yolo --skip-trust` instead of `--json -s read-only`
+ *   (`--skip-trust` required for headless/untrusted cwds; see gemini trusted-folders docs)
  * - No temp HOME needed — Gemini discovers skills from `.agents/skills/` in cwd
  * - Message events are streamed with `delta: true` — must concatenate
  */
@@ -121,7 +122,9 @@ export async function runGeminiSkill(opts: {
   }
 
   // Build gemini command
-  const args = ['-p', prompt, '--output-format', 'stream-json', '--yolo'];
+  // --skip-trust: headless/CI and temp cwds aren't in ~/.gemini/trustedFolders.json;
+  // without it gemini exits FatalUntrustedWorkspaceError before any model call.
+  const args = ['-p', prompt, '--output-format', 'stream-json', '--yolo', '--skip-trust'];
 
   // Spawn gemini — uses real HOME for auth (~/.gemini; HOME is allowlisted),
   // cwd for skill discovery. Hermetic scrub with gemini's auth surface

test/helpers/providers/gemini.ts

@@ -96,10 +96,17 @@ export function resultFromGeminiStream(
 /**
  * Gemini adapter — wraps the `gemini` CLI.
  *
- * Gemini CLI auth comes from either ~/.config/gemini/ or GOOGLE_API_KEY. Output
+ * Auth: GEMINI_API_KEY / GOOGLE_API_KEY (preferred), or ~/.gemini oauth.
- * format is NDJSON with `message`/`tool_use`/`result` events when `--output-format
+ *   Personal OAuth free-tier is no longer supported by gemini CLI — use an
- * stream-json` is requested. This adapter uses a single-response form for simplicity
+ *   AI Studio API key. Antigravity is a separate product/quota path.
- * in benchmarks; richer streaming lives in gemini-session-runner.ts.
+ *
+ * Headless flags always passed:
+ *   --output-format stream-json  — NDJSON events (message/tool_use/result)
+ *   --yolo                       — auto-approve tools (non-interactive)
+ *   --skip-trust                 — trust cwd for this session; required when
+ *                                  workdir is a temp/untrusted folder (benchmarks
+ *                                  use mkdtemp). Without it headless gemini exits
+ *                                  before calling the model.
  */
 export class GeminiAdapter implements ProviderAdapter {
   readonly name = 'gemini';
@@ -129,8 +136,10 @@ export class GeminiAdapter implements ProviderAdapter {
   async run(opts: RunOpts): Promise<RunResult> {
     const start = Date.now();
     // Default to --yolo (non-interactive) and stream-json output so we can parse
-    // tokens + tool calls. Callers can override via extraArgs.
+    // tokens + tool calls. --skip-trust is required for headless/temp workdirs
-    const args = ['-p', opts.prompt, '--output-format', 'stream-json', '--yolo'];
+    // (gemini CLI otherwise exits: "not running in a trusted directory").
+    // Callers can override via extraArgs.
+    const args = ['-p', opts.prompt, '--output-format', 'stream-json', '--yolo', '--skip-trust'];
     if (opts.model) args.push('--model', opts.model);
     if (opts.extraArgs) args.push(...opts.extraArgs);