fix(results): retry interrupted direct result pushes

Author: christso

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

@@ -985,6 +985,39 @@ export async function createDraftResultsPr(params: {
 
 const DIRECT_PUSH_MAX_RETRIES = 3;
 
+async function hasUnpushedCommits(repoDir: string, baseBranch: string): Promise<boolean> {
+  const { stdout } = await runGit(['rev-list', '--count', `origin/${baseBranch}..HEAD`], {
+    cwd: repoDir,
+    check: false,
+  });
+  return Number.parseInt(stdout.trim(), 10) > 0;
+}
+
+async function pushDirectResultsToBase(params: {
+  readonly normalized: Required<ResultsConfig>;
+  readonly repoDir: string;
+  readonly baseBranch: string;
+}): Promise<void> {
+  for (let attempt = 1; attempt <= DIRECT_PUSH_MAX_RETRIES; attempt++) {
+    try {
+      await runGit(['push', 'origin', `HEAD:${params.baseBranch}`], { cwd: params.repoDir });
+      updateStatusFile(params.normalized, {
+        last_synced_at: new Date().toISOString(),
+        last_error: undefined,
+      });
+      return;
+    } catch (error) {
+      const message = error instanceof Error ? error.message : String(error);
+      if (attempt < DIRECT_PUSH_MAX_RETRIES && message.includes('non-fast-forward')) {
+        await fetchResultsRepo(params.repoDir);
+        await runGit(['rebase', `origin/${params.baseBranch}`], { cwd: params.repoDir });
+      } else {
+        throw error;
+      }
+    }
+  }
+}
+
 /**
  * Push results directly to the base branch of the results repo.
  * Handles non-fast-forward conflicts by fetching, rebasing, and retrying.
@@ -1020,33 +1053,23 @@ export async function directPushResults(params: {
     check: false,
   });
   if (status.trim().length === 0) {
+    if (await hasUnpushedCommits(repoDir, baseBranch)) {
+      const aheadPaths = await getAheadPaths(repoDir, `origin/${baseBranch}`);
+      if (!areSafeResultsRepoPaths(aheadPaths)) {
+        throw new Error('Results repo has non-results committed changes');
+      }
+      await pushDirectResultsToBase({ normalized, repoDir, baseBranch });
+      return true;
+    }
     return false;
   }
 
   await runGit(['commit', '-m', params.commitMessage, '-m', `Agentv-Run: ${targetRunId}`], {
     cwd: repoDir,
   });
 
-  for (let attempt = 1; attempt <= DIRECT_PUSH_MAX_RETRIES; attempt++) {
-    try {
-      await runGit(['push', 'origin', `HEAD:${baseBranch}`], { cwd: repoDir });
-      updateStatusFile(normalized, {
-        last_synced_at: new Date().toISOString(),
-        last_error: undefined,
-      });
-      return true;
-    } catch (error) {
-      const message = error instanceof Error ? error.message : String(error);
-      if (attempt < DIRECT_PUSH_MAX_RETRIES && message.includes('non-fast-forward')) {
-        await fetchResultsRepo(repoDir);
-        await runGit(['rebase', `origin/${baseBranch}`], { cwd: repoDir });
-      } else {
-        throw error;
-      }
-    }
-  }
-
-  return false;
+  await pushDirectResultsToBase({ normalized, repoDir, baseBranch });
+  return true;
 }
 
 export interface GitListedRun {

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

@@ -1,5 +1,5 @@
 import { execSync } from 'node:child_process';
-import { mkdirSync, mkdtempSync, readFileSync, rmSync, writeFileSync } from 'node:fs';
+import { chmodSync, mkdirSync, mkdtempSync, readFileSync, rmSync, writeFileSync } from 'node:fs';
 import os from 'node:os';
 import path from 'node:path';
 
@@ -308,6 +308,56 @@ describe('results repo write path', () => {
     rmSync(rootDir, { recursive: true, force: true });
   });
 
+  it('retries an interrupted direct push without dropping the committed run', async () => {
+    const { remoteDir } = initializeRemoteRepo(rootDir);
+    const cloneDir = path.join(rootDir, 'results-clone');
+    const sourceDir = path.join(rootDir, 'source-run');
+    const runTimestamp = '2026-05-22T11-00-00-000Z';
+    const destinationPath = path.join('retry', runTimestamp);
+    const config = createResultsConfig(remoteDir, cloneDir);
+    const hookPath = path.join(remoteDir, 'hooks', 'pre-receive');
+    writeRunArtifacts(sourceDir, 'retry', '2026-05-22T11:00:00.000Z');
+
+    await ensureResultsRepoClone(config);
+    git('git config user.email "test@example.com"', cloneDir);
+    git('git config user.name "Test User"', cloneDir);
+
+    writeFileSync(hookPath, '#!/usr/bin/env sh\necho "simulated interrupted push" >&2\nexit 1\n');
+    chmodSync(hookPath, 0o755);
+
+    await expect(
+      directPushResults({
+        config,
+        sourceDir,
+        destinationPath,
+        commitMessage: 'feat(results): retry - 1/1 PASS (1.000)',
+      }),
+    ).rejects.toThrow(/simulated interrupted push/);
+    expect(git('git rev-list --count origin/main..HEAD', cloneDir)).toBe('1');
+    expect(git(`git --git-dir "${remoteDir}" ls-tree -r --name-only main`, rootDir)).not.toContain(
+      `.agentv/results/runs/retry/${runTimestamp}/benchmark.json`,
+    );
+
+    rmSync(hookPath, { force: true });
+
+    await expect(
+      directPushResults({
+        config,
+        sourceDir,
+        destinationPath,
+        commitMessage: 'feat(results): retry - 1/1 PASS (1.000)',
+      }),
+    ).resolves.toBe(true);
+
+    expect(git('git rev-list --count origin/main..HEAD', cloneDir)).toBe('0');
+    expect(git(`git --git-dir "${remoteDir}" ls-tree -r --name-only main`, rootDir)).toContain(
+      `.agentv/results/runs/retry/${runTimestamp}/benchmark.json`,
+    );
+    expect(git(`git --git-dir "${remoteDir}" log -1 --pretty=%B main`, rootDir)).toContain(
+      `Agentv-Run: retry::${runTimestamp}`,
+    );
+  }, 20000);
+
   it('commits pushed runs into the configured clone with an Agentv-Run trailer', async () => {
     const { remoteDir } = initializeRemoteRepo(rootDir);
     const cloneDir = path.join(rootDir, 'results-clone');