togethercomputer · blainekasten · May 12, 2026 · May 12, 2026 · May 12, 2026 · May 12, 2026
diff --git a/.github/workflows/sync-vercel-detect-agent-upstream.yml b/.github/workflows/sync-vercel-detect-agent-upstream.yml
@@ -0,0 +1,205 @@
+# Sync Python port with upstream Vercel `packages/detect-agent` when it changes.
+#
+# Setup:
+# - Repository secret: CURSOR_API_KEY (https://github.com/marketplace/actions/cursor-action)
+# - No GitHub PAT: the final `gh` step uses the built-in GITHUB_TOKEN (see permissions below).
+# - Uses actions/cache to remember the last synced upstream commit for that path.
+# - First successful run bootstraps cache only (no agent); the next drift runs the agent + publish steps.
+# - Cursor only edits files and runs checks; this workflow commits, pushes, and opens/updates the PR.
+#
+# Test on your PR (same repo only; fork PRs are ignored):
+# 1. Create a repo label named `test-upstream-sync` (Settings -> Labels) if it does not exist.
+# 2. Open your PR, add the `test-upstream-sync` label — this workflow runs once (labeled event).
+# 3. PR runs use a separate Actions cache key so they do not overwrite the mainline upstream SHA.
+# 4. Remove the label when done (optional).
+#
+# Manual run: Actions -> "Sync Vercel detect-agent upstream" -> Run workflow (pick your branch to test workflow_dispatch changes).
+
+name: Sync Vercel detect-agent upstream
+
+on:
+  schedule:
+    # Daily 14:05 UTC — adjust as you like
+    - cron: "5 14 * * *"
+  workflow_dispatch:
+  pull_request:
+    types: [labeled]
+
+concurrency:
+  group: sync-vercel-detect-agent-${{ github.ref }}
+  cancel-in-progress: false
+
+jobs:
+  sync:
+    if: >-
+      github.event_name != 'pull_request' ||
+      (github.event_name == 'pull_request' &&
+       github.event.label.name == 'test-upstream-sync' &&
+       github.event.pull_request.head.repo.full_name == github.repository)
+    runs-on: ubuntu-latest
+    permissions:
+      contents: write
+      pull-requests: write
+    env:
+      # runner.* is not allowed in job-level env; use stable paths (fresh VM per hosted job).
+      UPSTREAM_STATE_DIR: /tmp/detect-agent-upstream-state
+      UPSTREAM_VERCEL_MIRROR: /tmp/vercel-detect-agent-bare.git
+      UPSTREAM_DIFF_FILE: /tmp/vercel-packages-detect-agent.diff
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          persist-credentials: true
+          ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
+
+      - uses: actions/cache@v4
+        id: vercel_mirror_cache
+        with:
+          path: /tmp/vercel-detect-agent-bare.git
+          key: vercel-github-bare-mirror-v1-${{ runner.os }}
+
+      - uses: actions/cache@v4
+        id: upstream_state_cache
+        with:
+          path: /tmp/detect-agent-upstream-state
+          key: vercel-detect-agent-upstream-state-v1-${{ runner.os }}-${{ github.event_name == 'pull_request' && format('pr-{0}', github.event.pull_request.number) || 'default' }}
+
+      - name: Compute upstream drift
+        id: drift
+        shell: bash
+        run: ./scripts/vercel_upstream_drift.sh
+
+      - name: Prepare default branch for sync
+        if: steps.drift.outputs.should_sync == 'true'
+        env:
+          DEFAULT_BRANCH: ${{ github.event.repository.default_branch }}
+        run: |
+          set -euo pipefail
+          db="${DEFAULT_BRANCH:-main}"
+          git fetch origin
+          git checkout "$db"
+          git pull --ff-only "origin/$db"
+
+      - name: Configure git committer (actions bot)
+        if: steps.drift.outputs.should_sync == 'true'
+        run: |
+          set -euo pipefail
+          git config user.name "github-actions[bot]"
+          git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
+
+      - uses: astral-sh/setup-uv@v5
+        if: steps.drift.outputs.should_sync == 'true'
+        with:
+          enable-cache: true
+          python-version: "3.9"
+
+      - name: Install Python deps (uv)
+        if: steps.drift.outputs.should_sync == 'true'
+        run: uv sync --extra dev
+
+      - name: Run Cursor agent (port + verify)
+        id: cursor
+        if: steps.drift.outputs.should_sync == 'true'
+        uses: PunGrumpy/cursor-action@v1.0.0
+        with:
+          api-key: ${{ secrets.CURSOR_API_KEY }}
+          model: composer-2
+          working-directory: .
+          permissions: read-write
+          timeout: "3600"
+          prompt: |
+            You are updating this repository's Python port of Vercel's `packages/detect-agent` to match upstream changes.
+
+            The workflow has already checked out the default branch and pulled latest; stay on the current branch. Do not run any `git` commands and do not use `gh`.
+
+            Upstream facts:
+            - merge_base: ${{ steps.drift.outputs.merge_base }}
+            - last_recorded_upstream_tip: ${{ steps.drift.outputs.last_recorded }}
+            - new_upstream_tip: ${{ steps.drift.outputs.new_sha }}
+            - compare URL: ${{ steps.drift.outputs.compare_url }}
+
+            Read the unified diff from this absolute path. Do not edit that path; it is CI scratch only:
+            ${{ steps.drift.outputs.diff_path }}
+
+            Your job:
+            1. Port upstream logic into `detect_agent/__init__.py` and `tests/` as needed. Mirror detection order and semantics from the TypeScript sources; keep typing style consistent with the existing Python file.
+            2. Update `README.md` / `CHANGELOG.md` here if upstream docs/changelog warrant it.
+            3. Run: `uv sync --extra dev && uv run pytest && uv run ruff check . && uv run ruff format --check .`
+
+            If the diff file is empty but the SHAs differ, use the compare URL and current upstream `main` tree for that package to align the port.
+
+      - name: Commit, push, and open or update PR
+        id: publish
+        if: steps.drift.outputs.should_sync == 'true' && steps.cursor.outcome == 'success'
+        env:
+          GH_TOKEN: ${{ github.token }}
+          DEFAULT_BRANCH: ${{ github.event.repository.default_branch }}
+          MERGE_BASE: ${{ steps.drift.outputs.merge_base }}
+          NEW_SHA: ${{ steps.drift.outputs.new_sha }}
+          COMPARE_URL: ${{ steps.drift.outputs.compare_url }}
+          BRANCH_SLUG: ${{ steps.drift.outputs.branch_slug }}
+        shell: bash
+        run: |
+          set -euo pipefail
+          db="${DEFAULT_BRANCH:-main}"
+          BRANCH_NAME="sync/vercel-detect-agent-${BRANCH_SLUG}"
+
+          git branch -D "$BRANCH_NAME" 2>/dev/null || true
+          git checkout -b "$BRANCH_NAME"
+
+          git add detect_agent tests README.md CHANGELOG.md pyproject.toml uv.lock 2>/dev/null || true
+
+          if git diff --staged --quiet; then
+            echo "::notice::No staged changes after agent; skipping commit and PR."
+            exit 0
+          fi
+
+          short_mb="${MERGE_BASE:0:7}"
+          short_new="${NEW_SHA:0:7}"
+          git commit -m "sync: port vercel detect-agent ${short_mb}..${short_new}"
+
+          git push -u origin "$BRANCH_NAME" --force
+
+          EXISTING_PR="$(gh pr list --state open --head "$BRANCH_NAME" --json number --jq '.[0].number // empty' 2>/dev/null || true)"
+
+          PR_TITLE="sync: port vercel detect-agent (${short_mb}..${short_new})"
+          PR_BODY="$(printf '%s\n\n%s\n\n%s\n' "Ports [\`packages/detect-agent\`](https://github.com/vercel/vercel/tree/main/packages/detect-agent) from [vercel/vercel](https://github.com/vercel/vercel)." "Compare: ${COMPARE_URL}" "cc @blainekasten")"
+
+          if [ -n "${EXISTING_PR}" ]; then
+            gh pr edit "${EXISTING_PR}" --title "${PR_TITLE}" --body "${PR_BODY}"
+            gh pr edit "${EXISTING_PR}" --add-assignee blainekasten 2>/dev/null || true
+            echo "Updated existing PR #${EXISTING_PR}"
+          else
+            gh pr create \
+              --title "${PR_TITLE}" \
+              --body "${PR_BODY}" \
+              --base "${db}" \
+              --head "${BRANCH_NAME}" \
+              --assignee blainekasten
+          fi
+
+      - name: Record synced upstream SHA
+        if: success() && steps.drift.outputs.should_sync == 'true' && steps.cursor.outcome == 'success' && steps.publish.outcome == 'success'
+        shell: bash
+        env:
+          NEW_SHA: ${{ steps.drift.outputs.new_sha }}
+        run: |
+          set -euo pipefail
+          mkdir -p "${UPSTREAM_STATE_DIR}"
+          printf '%s\n' "${NEW_SHA}" >"${UPSTREAM_STATE_DIR}/last_packages_detect_agent_commit.txt"
+
+      - name: Job summary
+        if: always()
+        shell: bash
+        run: |
+          set -euo pipefail
+          {
+            echo "## Upstream sync"
+            echo "- should_sync: ${{ steps.drift.outputs.should_sync }}"
+            echo "- new_sha: ${{ steps.drift.outputs.new_sha }}"
+            echo "- compare: ${{ steps.drift.outputs.compare_url }}"
+          } >>"${GITHUB_STEP_SUMMARY}"
+          if [[ "${{ steps.drift.outputs.should_sync }}" == "true" ]]; then
+            echo "- cursor exit: ${{ steps.cursor.outputs.exit-code }}" >>"${GITHUB_STEP_SUMMARY}"
+            echo "- publish outcome: ${{ steps.publish.outcome }}" >>"${GITHUB_STEP_SUMMARY}"
+          fi
diff --git a/scripts/vercel_upstream_drift.sh b/scripts/vercel_upstream_drift.sh
@@ -0,0 +1,107 @@
+#!/usr/bin/env bash
+# Compute whether vercel/vercel changed packages/detect-agent since the last run.
+# Intended for GitHub Actions; optionally writes key=value lines to GITHUB_OUTPUT.
+#
+# Required env:
+#   UPSTREAM_STATE_DIR     - directory holding last_packages_detect_agent_commit.txt
+#   UPSTREAM_VERCEL_MIRROR - path to (or parent of) bare clone of github.com/vercel/vercel.git
+#   UPSTREAM_DIFF_FILE     - where to write unified diff when drift is detected
+#
+# Optional:
+#   GITHUB_OUTPUT          - when set, writes should_sync, new_sha, merge_base, last_recorded,
+#                            diff_path, compare_url (booleans as true/false strings)
+
+set -euo pipefail
+
+UPSTREAM_STATE_DIR="${UPSTREAM_STATE_DIR:?UPSTREAM_STATE_DIR is required}"
+UPSTREAM_VERCEL_MIRROR="${UPSTREAM_VERCEL_MIRROR:?UPSTREAM_VERCEL_MIRROR is required}"
+UPSTREAM_DIFF_FILE="${UPSTREAM_DIFF_FILE:?UPSTREAM_DIFF_FILE is required}"
+
+path='packages/detect-agent'
+LAST_SHA_FILE="${UPSTREAM_STATE_DIR}/last_packages_detect_agent_commit.txt"
+
+append_kv() {
+  if [[ -z "${GITHUB_OUTPUT:-}" ]]; then
+    return 0
+  fi
+  printf '%s=%s\n' "$1" "$2" >>"$GITHUB_OUTPUT"
+}
+
+mkdir -p "$UPSTREAM_STATE_DIR"
+
+if [[ ! -d "$UPSTREAM_VERCEL_MIRROR" ]]; then
+  git clone --mirror https://github.com/vercel/vercel.git "$UPSTREAM_VERCEL_MIRROR"
+fi
+
+git -C "$UPSTREAM_VERCEL_MIRROR" fetch --prune origin
+
+# Mirror clones expose the default branch as refs/heads/main, not origin/main.
+main_ref="main"
+if ! git -C "$UPSTREAM_VERCEL_MIRROR" rev-parse -q --verify "${main_ref}^{commit}" >/dev/null 2>&1; then
+  main_ref="origin/main"
+fi
+if ! git -C "$UPSTREAM_VERCEL_MIRROR" rev-parse -q --verify "${main_ref}^{commit}" >/dev/null 2>&1; then
+  echo "failed to resolve main branch tip (tried main, origin/main) in ${UPSTREAM_VERCEL_MIRROR}" >&2
+  exit 1
+fi
+
+NEW_SHA="$(git -C "$UPSTREAM_VERCEL_MIRROR" log -1 --format=%H "$main_ref" -- "$path")"
+if [[ -z "$NEW_SHA" ]]; then
+  echo "failed to resolve latest commit for $path on ${main_ref}" >&2
+  exit 1
+fi
+
+if [[ ! -f "$LAST_SHA_FILE" ]]; then
+  printf '%s\n' "$NEW_SHA" >"$LAST_SHA_FILE"
+  echo "bootstrapped upstream state (no sync): $NEW_SHA" >&2
+  append_kv should_sync false
+  append_kv new_sha "$NEW_SHA"
+  exit 0
+fi
+
+LAST_SHA="$(tr -d '[:space:]' <"$LAST_SHA_FILE" || true)"
+if [[ -z "$LAST_SHA" ]]; then
+  printf '%s\n' "$NEW_SHA" >"$LAST_SHA_FILE"
+  echo "repaired empty state file -> $NEW_SHA" >&2
+  append_kv should_sync false
+  append_kv new_sha "$NEW_SHA"
+  exit 0
+fi
+
+if [[ "$LAST_SHA" == "$NEW_SHA" ]]; then
+  echo "no upstream drift: $NEW_SHA" >&2
+  append_kv should_sync false
+  append_kv new_sha "$NEW_SHA"
+  exit 0
+fi
+
+echo "upstream drift: $LAST_SHA -> $NEW_SHA" >&2
+
+if ! git -C "$UPSTREAM_VERCEL_MIRROR" cat-file -e "${LAST_SHA}^{commit}" 2>/dev/null; then
+  echo "widening fetch for stored sha" >&2
+  git -C "$UPSTREAM_VERCEL_MIRROR" fetch --prune origin "${LAST_SHA}" || true
+fi
+
+if ! git -C "$UPSTREAM_VERCEL_MIRROR" cat-file -e "${LAST_SHA}^{commit}" 2>/dev/null; then
+  echo "stored sha $LAST_SHA not found; delete ${LAST_SHA_FILE} to re-bootstrap" >&2
+  exit 1
+fi
+
+MB="$(git -C "$UPSTREAM_VERCEL_MIRROR" merge-base "$LAST_SHA" "$NEW_SHA" 2>/dev/null || true)"
+if [[ -z "$MB" ]]; then
+  MB="$LAST_SHA"
+fi
+
+git -C "$UPSTREAM_VERCEL_MIRROR" diff "${MB}..${NEW_SHA}" -- "$path" >"$UPSTREAM_DIFF_FILE" || true
+
+COMPARE_URL="https://github.com/vercel/vercel/compare/${MB}...${NEW_SHA}"
+
+append_kv should_sync true
+append_kv new_sha "$NEW_SHA"
+append_kv branch_slug "${NEW_SHA:0:7}"
+append_kv merge_base "$MB"
+append_kv last_recorded "$LAST_SHA"
+append_kv diff_path "$UPSTREAM_DIFF_FILE"
+append_kv compare_url "$COMPARE_URL"
+
+exit 0