feat(results): direct-push auto_push to base branch

apps/cli/src/commands/results/remote.ts

@@ -5,14 +5,11 @@ import {
   type EvaluationResult,
   type ResultsExportConfig,
   type ResultsRepoStatus,
-  commitAndPushResultsBranch,
-  createDraftResultsPr,
+  directPushResults,
   directorySizeBytes,
   getResultsRepoStatus,
   loadConfig,
-  prepareResultsRepoBranch,
   resolveResultsRepoRunsDir,
-  stageResultsArtifacts,
   syncResultsRepo,
 } from '@agentv/core';
 
@@ -78,15 +75,6 @@ function statusForResult(result: EvaluationResult): 'PASS' | 'FAIL' | 'ERROR' {
   return result.score >= DEFAULT_THRESHOLD ? 'PASS' : 'FAIL';
 }
 
-function slugify(value: string): string {
-  return value
-    .trim()
-    .replace(/[^A-Za-z0-9._/-]+/g, '-')
-    .replace(/\/+/g, '/')
-    .replace(/^-+|-+$/g, '')
-    .slice(0, 120);
-}
-
 function getRelativeRunPath(cwd: string, runDir: string): string {
   const relative = path.relative(path.join(cwd, '.agentv', 'results', 'runs'), runDir);
   if (!relative.startsWith('..') && !path.isAbsolute(relative)) {
@@ -98,23 +86,6 @@ function getRelativeRunPath(cwd: string, runDir: string): string {
   return experiment && experiment !== runName ? path.join(experiment, runName) : runName;
 }
 
-function buildBranchName(
-  config: Required<ResultsExportConfig>,
-  payload: RemoteExportPayload,
-): string {
-  const timestamp = path.basename(payload.run_dir);
-  const evalStem =
-    payload.test_files.length === 1
-      ? path
-          .basename(payload.test_files[0])
-          .replace(/\.eval\.ya?ml$/i, '')
-          .replace(/\.[^.]+$/i, '')
-      : `${payload.test_files.length}-evals`;
-  const experiment = slugify(payload.experiment ?? 'default');
-  const branchLeaf = slugify(`${experiment}-${evalStem}-${timestamp}`) || timestamp;
-  return `${config.branch_prefix}/${branchLeaf}`;
-}
-
 function buildCommitTitle(payload: RemoteExportPayload): string {
   const passed = payload.results.filter((result) => result.score >= DEFAULT_THRESHOLD).length;
   const avgScore =
@@ -125,35 +96,6 @@ function buildCommitTitle(payload: RemoteExportPayload): string {
   return `feat(results): ${experiment} - ${passed}/${payload.results.length} PASS (${avgScore.toFixed(3)})`;
 }
 
-function buildPrBody(payload: RemoteExportPayload): string {
-  const sections = payload.eval_summaries
-    .map((summary) => {
-      const table = summary.results
-        .map((result) => `| ${result.test_id} | ${result.score.toFixed(3)} | ${result.status} |`)
-        .join('\n');
-      return [
-        `### ${summary.eval_file}`,
-        '',
-        `Summary: ${summary.passed}/${summary.total} PASS (${summary.avg_score.toFixed(3)})`,
-        '',
-        '| Test | Score | Status |',
-        '|---|---|---|',
-        table || '| (no results) | 0.000 | ERROR |',
-      ].join('\n');
-    })
-    .join('\n\n');
-
-  return [
-    '## Results',
-    '',
-    sections,
-    '',
-    `Run: ${path.basename(payload.run_dir)}`,
-    `Experiment: ${payload.experiment ?? 'default'}`,
-    `Eval Files: ${payload.test_files.join(', ')}`,
-  ].join('\n');
-}
-
 async function maybeWarnLargeArtifact(runDir: string): Promise<void> {
   const sizeBytes = await directorySizeBytes(runDir);
   if (sizeBytes > SIZE_WARNING_BYTES) {
@@ -279,43 +221,22 @@ export async function maybeAutoExportRunArtifacts(payload: RemoteExportPayload):
   try {
     await maybeWarnLargeArtifact(payload.run_dir);
 
-    const branchName = buildBranchName(config, payload);
-    const prepared = await prepareResultsRepoBranch(config, branchName);
-
-    try {
-      const relativeRunPath = getRelativeRunPath(payload.cwd, payload.run_dir);
-      const destinationDir = path.join(prepared.repoDir, config.path, relativeRunPath);
-      await stageResultsArtifacts({
-        repoDir: prepared.repoDir,
-        sourceDir: payload.run_dir,
-        destinationDir,
-      });
-
-      const commitTitle = buildCommitTitle(payload);
-      const changed = await commitAndPushResultsBranch({
-        repoDir: prepared.repoDir,
-        branchName,
-        commitMessage: commitTitle,
-      });
-
-      if (!changed) {
-        console.warn('Warning: results export produced no git changes. Skipping PR creation.');
-        return;
-      }
-
-      const prUrl = await createDraftResultsPr({
-        repo: config.repo,
-        repoDir: prepared.repoDir,
-        baseBranch: prepared.baseBranch,
-        branchName,
-        title: commitTitle,
-        body: buildPrBody(payload),
-      });
-
-      console.log(`Remote results draft PR created: ${prUrl}`);
-    } finally {
-      await prepared.cleanup();
+    const relativeRunPath = getRelativeRunPath(payload.cwd, payload.run_dir);
+    const commitTitle = buildCommitTitle(payload);
+
+    const pushed = await directPushResults({
+      config,
+      sourceDir: payload.run_dir,
+      destinationPath: relativeRunPath,
+      commitMessage: commitTitle,
+    });
+
+    if (!pushed) {
+      console.warn('Warning: results export produced no git changes. Skipping push.');
+      return;
     }
+
+    console.log(`Results pushed to ${config.repo} (${config.path}/${relativeRunPath})`);
   } catch (error) {
     console.warn(`Warning: skipping results export: ${getStatusMessage(error)}`);
     console.warn("Warning: Run 'gh auth login' if GitHub authentication is missing.");

apps/web/src/content/docs/docs/tools/studio.mdx

@@ -203,11 +203,12 @@ results:
   export:
     repo: EntityProcess/agentv-evals   # GitHub repo (owner/repo or full URL)
     path: runs                          # Directory within the repo
-    auto_push: true                     # Push automatically after every eval run
-    branch_prefix: eval-results         # Branch naming prefix (default: eval-results)
+    auto_push: true                     # Push directly to base branch after every eval run
 ```
 
-With `auto_push: true`, every `agentv eval run` or `agentv pipeline bench` automatically creates a draft PR in the configured repo with a structured results table.
+With `auto_push: true`, every `agentv eval` or `agentv pipeline bench` pushes results directly to the configured repo's base branch (e.g., `main`). Results appear immediately in Studio without requiring PR merges.
+
+Each run writes to a unique timestamped directory, so concurrent pushes from multiple machines are safe — non-fast-forward conflicts are resolved automatically via rebase retry.
 
 ### Authentication
 

packages/core/src/evaluation/results-repo.ts

@@ -396,3 +396,60 @@ export async function createDraftResultsPr(params: {
   );
   return stdout.trim();
 }
+
+const DIRECT_PUSH_MAX_RETRIES = 3;
+
+/**
+ * Push results directly to the base branch of the results repo.
+ * Handles non-fast-forward conflicts by pulling with rebase and retrying.
+ * Returns true if artifacts were pushed, false if no changes were detected.
+ */
+export async function directPushResults(params: {
+  readonly config: ResultsExportConfig;
+  readonly sourceDir: string;
+  readonly destinationPath: string;
+  readonly commitMessage: string;
+}): Promise<boolean> {
+  const normalized = normalizeResultsExportConfig(params.config);
+  const repoDir = await ensureResultsRepoClone(normalized);
+  const baseBranch = await resolveDefaultBranch(repoDir);
+  await updateCacheRepo(repoDir, baseBranch);
+
+  const destinationDir = path.join(repoDir, normalized.path, params.destinationPath);
+  await stageResultsArtifacts({
+    repoDir,
+    sourceDir: params.sourceDir,
+    destinationDir,
+  });
+
+  await runGit(['add', '--all'], { cwd: repoDir });
+  const { stdout: status } = await runGit(['status', '--porcelain'], {
+    cwd: repoDir,
+    check: false,
+  });
+  if (status.trim().length === 0) {
+    return false;
+  }
+
+  await runGit(['commit', '-m', params.commitMessage], { cwd: repoDir });
+
+  for (let attempt = 1; attempt <= DIRECT_PUSH_MAX_RETRIES; attempt++) {
+    try {
+      await runGit(['push', 'origin', baseBranch], { cwd: repoDir });
+      updateStatusFile(normalized, {
+        last_synced_at: new Date().toISOString(),
+        last_error: undefined,
+      });
+      return true;
+    } catch (error) {
+      const message = error instanceof Error ? error.message : String(error);
+      if (attempt < DIRECT_PUSH_MAX_RETRIES && message.includes('non-fast-forward')) {
+        await runGit(['pull', '--rebase', 'origin', baseBranch], { cwd: repoDir });
+      } else {
+        throw error;
+      }
+    }
+  }
+
+  return false;
+}

packages/core/src/index.ts

@@ -73,6 +73,7 @@ export {
   commitAndPushResultsBranch,
   pushResultsRepoBranch,
   createDraftResultsPr,
+  directPushResults,
   type CheckedOutResultsRepoBranch,
   type PreparedResultsRepoBranch,
   type ResultsRepoCachePaths,