diff --git a/.github/workflows/auto-merge.yml b/.github/workflows/auto-merge.yml
new file mode 100644
index 0000000..a480aaf
--- /dev/null
+++ b/.github/workflows/auto-merge.yml
@@ -0,0 +1,73 @@
+# Auto-merge — nobody is in the merge loop.
+#
+# Green, ready PRs merge themselves and deploy themselves. The owner does not
+# review PRs, and background-job agent sessions are barred from merging by hand,
+# so the policy lives in scripts/ci/auto-merge-sweep.sh (read it — it defines
+# exactly what "ready" means and how a PR is held back).
+#
+# Two triggers, deliberately:
+#   workflow_run — merges within seconds of CI going green (the common path).
+#   schedule     — a safety net. Catches PRs whose checks finished while this
+#                  workflow was failing/disabled, and PRs whose last check was
+#                  an external status that reported after CI. Without it, a PR
+#                  that went green "off-cycle" waits forever.
+#
+# To stop all of this: delete this file, or add a `hold` label to a PR.
+
+name: Auto-merge
+
+on:
+  workflow_run:
+    workflows: ['CI']
+    types: [completed]
+  # HOURLY, not */10 as in the rest of the fleet — this repo is PRIVATE, so
+  # Actions minutes are billed. Every other repo running this sweep is public,
+  # where scheduled runs are free. At */10 the safety net alone would burn
+  # ~4,300 billed minutes a month (144 runs/day, each rounded up to a minute)
+  # against a 2,000-minute free allowance, and it would spend most of them
+  # discovering there are no PRs. Hourly costs ~720.
+  #
+  # This barely slows anything down: the schedule is only the safety net. The
+  # common path is workflow_run above, which fires within seconds of CI going
+  # green. Make this repo public and */10 becomes free again.
+  schedule:
+    - cron: '0 * * * *'
+  workflow_dispatch: {}
+
+permissions:
+  contents: write # merge the PR
+  pull-requests: write # read PR state, delete the branch
+  actions: write # dispatch the re-arm workflows
+  # Required because THIS repo is private. The sweep reads statusCheckRollup to
+  # decide whether a PR is green; a public repo answers that with no explicit
+  # scope, which is why the copy of this file going around the fleet omits these
+  # two lines and still works in ~20 public repos. On a private repo the same
+  # query fails with 'Resource not accessible by integration' and the sweep
+  # merges nothing, ever — see ivy-portal, where that went unnoticed because a
+  # repo that merges nothing looks like a repo with nothing to merge.
+  checks: read # check-run conclusions (CI jobs)
+  statuses: read # commit statuses (external reporters)
+
+# Never let two sweeps merge concurrently — they would race on the same PRs.
+concurrency:
+  group: auto-merge
+  cancel-in-progress: false
+
+jobs:
+  sweep:
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Merge every green, ready PR
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          GH_REPO: ${{ github.repository }}
+          BASE_BRANCH: ${{ github.event.repository.default_branch }}
+          CI_WORKFLOW: ci.yml
+          # Everything that runs on push and therefore would NOT fire after a
+          # merge made with the default GITHUB_TOKEN. Keep this in sync when a
+          # push-triggered workflow is added.
+          REARM_WORKFLOWS: ci.yml
+        run: bash scripts/ci/auto-merge-sweep.sh
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
new file mode 100644
index 0000000..61c4430
--- /dev/null
+++ b/.github/workflows/ci.yml
@@ -0,0 +1,69 @@
+name: CI
+
+# Nothing has ever run this repo's checks. `npm run verify` (typecheck + lint +
+# vitest) existed from the seed commit and was only ever run by hand, which for
+# an agent-built project means: not run. This is the gate that makes it real,
+# and it is also what lets auto-merge work at all — the sweep refuses to merge a
+# PR with zero checks, so without CI nothing here could ever ship itself.
+
+on:
+  # Load-bearing for the merge train: a merge made by auto-merge.yml uses the
+  # default GITHUB_TOKEN, and a push made with that token does NOT trigger
+  # workflows. The sweep dispatches this explicitly afterwards to verify the
+  # merge on main; without this trigger that re-arm silently fails.
+  workflow_dispatch: {}
+  push:
+    branches: [main]
+  pull_request:
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  ci:
+    runs-on: ubuntu-latest
+
+    # `next build` needs a real database. The home page reads DATABASE_URL at
+    # module scope, so page-data collection dies with "DATABASE_URL is not set"
+    # on a bare runner — verified before writing this. A schema-only Postgres is
+    # enough: the build only needs the queries to run, not to return rows.
+    services:
+      postgres:
+        image: postgres:17
+        env:
+          POSTGRES_PASSWORD: ci
+          POSTGRES_DB: hamstercheek
+        ports: ['5432:5432']
+        options: >-
+          --health-cmd pg_isready --health-interval 5s
+          --health-timeout 5s --health-retries 10
+
+    env:
+      DATABASE_URL: postgres://postgres:ci@localhost:5432/hamstercheek
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Setup Node
+        uses: actions/setup-node@v4
+        with:
+          node-version: 22
+          cache: npm
+
+      - name: Install dependencies
+        run: npm ci
+
+      # ONE canonical bundle — the exact `npm run verify` you run locally, so a
+      # green local verify cannot diverge from a green CI. Needs no database:
+      # the unit tests are pure.
+      - name: Verify (typecheck + lint + test)
+        run: npm run verify
+
+      - name: Apply migrations (the build queries this DB)
+        run: npm run db:migrate
+
+      # Catches the class verify cannot: code that typechecks but fails to
+      # compile or blows up while Next collects page data.
+      - name: Build
+        run: npm run build
diff --git a/scripts/ci/auto-merge-sweep.sh b/scripts/ci/auto-merge-sweep.sh
new file mode 100755
index 0000000..08ae2db
--- /dev/null
+++ b/scripts/ci/auto-merge-sweep.sh
@@ -0,0 +1,208 @@
+#!/usr/bin/env bash
+#
+# Merge every open PR that is ready and fully green, then re-arm CI/CD.
+#
+# WHY THIS EXISTS
+# ---------------
+# Nobody reviews PRs on this fleet — the owner explicitly does not want to be in
+# the merge loop, and background-job agent sessions are barred from merging by
+# hand. So the policy lives here, in the repo, where it is visible, revocable,
+# and applies uniformly to every PR instead of depending on who opened it.
+#
+# THE POLICY
+#   merge a PR  <=>  it is not a draft
+#                    AND carries no hold label
+#                    AND has at least one check
+#                    AND every check has finished green
+#                    AND GitHub reports it cleanly mergeable
+#
+# Anything else is left alone for the next sweep. Nothing here forces a merge:
+# a red or pending PR simply waits, and a draft waits forever. To hold a ready
+# PR back, mark it a draft or add one of the hold labels below.
+#
+# ONE PR PER SWEEP, OLDEST FIRST, AND ONLY ONTO A GREEN BASE
+# ----------------------------------------------------------
+# A PR's checks prove *that PR against the base it branched from* — not against
+# the other PRs sitting next to it. Merging a batch in one pass would put a
+# combination onto the base that nothing ever built. So this script merges at
+# most one PR, then hands control back to CI: the merge train advances one car
+# per sweep, and every car is verified on the base before the next one couples.
+#
+# For the same reason it refuses to merge while the base's CI is red or still
+# running. Red base => stop adding changes until it is fixed; running CI => the
+# answer is not in yet. Both simply defer to the next sweep.
+#
+# THE RE-ARM (do not remove)
+#   A push made with the default GITHUB_TOKEN does NOT trigger workflows. Both
+#   CI and the deploy workflow here run on push, so a merge from this script
+#   would otherwise land on the base branch and never build or ship. Worse, the
+#   green-base guard above keys on "a CI run exists for the current tip" — with
+#   no CI run ever produced by an automated merge, the very next sweep would
+#   block forever. The explicit workflow_dispatch calls at the end restore both.
+#
+#   REARM_WORKFLOWS is set by .github/workflows/auto-merge.yml and lists exactly
+#   the workflows that would otherwise have fired on push.
+
+set -euo pipefail
+
+REPO="${GH_REPO:?GH_REPO must be set}"
+BASE_BRANCH="${BASE_BRANCH:-main}"
+CI_WORKFLOW="${CI_WORKFLOW:-ci.yml}"
+REARM_WORKFLOWS="${REARM_WORKFLOWS:-$CI_WORKFLOW}"
+
+# A PR wearing any of these is never merged automatically.
+HOLD_LABELS='["hold","no-automerge","do-not-merge","wip"]'
+
+echo "[auto-merge] sweeping open PRs against ${BASE_BRANCH} in ${REPO}"
+
+# Never add changes to a base that is red or mid-verification.
+#
+# The run has to belong to the CURRENT tip of the base branch. Checking only
+# "the latest CI run" is a trap: right after a merge, the newest run is still
+# the *previous* commit's — and it is green — so the guard would wave through a
+# second merge onto a commit nothing has verified yet. That is exactly the
+# batching this script exists to prevent.
+base_sha=$(gh api "repos/${REPO}/commits/${BASE_BRANCH}" --jq '.sha')
+base_ci=$(gh run list --repo "$REPO" --workflow "$CI_WORKFLOW" --branch "$BASE_BRANCH" --limit 1 \
+  --json status,conclusion,headSha --jq '.[0] // empty')
+
+if [ -z "$base_ci" ]; then
+  echo "[auto-merge] no CI history for ${BASE_BRANCH} — proceeding"
+else
+  base_status=$(printf '%s' "$base_ci" | jq -r '.status')
+  base_conclusion=$(printf '%s' "$base_ci" | jq -r '.conclusion // ""')
+  base_ci_sha=$(printf '%s' "$base_ci" | jq -r '.headSha')
+
+  if [ "$base_ci_sha" != "$base_sha" ]; then
+    echo "[auto-merge] ${BASE_BRANCH} is at ${base_sha:0:8} but the newest CI run is for ${base_ci_sha:0:8} — waiting for CI to catch up"
+    exit 0
+  fi
+  if [ "$base_status" != "completed" ]; then
+    echo "[auto-merge] ${BASE_BRANCH} CI is still running — deferring to the next sweep"
+    exit 0
+  fi
+  if [ "$base_conclusion" != "success" ]; then
+    echo "[auto-merge] ${BASE_BRANCH} CI is ${base_conclusion} — refusing to merge onto a broken base" >&2
+    exit 0
+  fi
+fi
+
+prs_json=$(gh pr list --repo "$REPO" --state open --base "$BASE_BRANCH" --limit 50 \
+  --json number,title,isDraft,mergeable,mergeStateStatus,labels,statusCheckRollup)
+
+count=$(printf '%s' "$prs_json" | jq 'length')
+if [ "$count" -eq 0 ]; then
+  echo "[auto-merge] no open PRs"
+  exit 0
+fi
+
+merged_any=0
+
+# OLDEST FIRST. `gh pr list` returns newest-first, and this loop merges the
+# first eligible PR and stops — so the newest green PR wins every sweep and an
+# older one can wait indefinitely. Observed in maonakamoto/fleetcrown on
+# 2026-08-06: consecutive sweeps merged the two newest PRs while three older
+# green ones were never even evaluated. With several agent sessions opening PRs
+# continuously, "newest wins" is starvation, and it starves the PR whose checks
+# were proven against the most now-stale base.
+#
+# PR numbers increase monotonically with creation, so sorting ascending is FIFO.
+for number in $(printf '%s' "$prs_json" | jq -r 'sort_by(.number) | .[].number'); do
+  pr=$(printf '%s' "$prs_json" | jq -c --argjson n "$number" '.[] | select(.number == $n)')
+  title=$(printf '%s' "$pr" | jq -r '.title')
+
+  # A rollup entry is either a CheckRun (status + conclusion) or a commit
+  # StatusContext (state) — external services report as the latter.
+  verdict=$(printf '%s' "$pr" | jq -r --argjson hold "$HOLD_LABELS" '
+    def ok:
+      if has("state") then (.state == "SUCCESS")
+      else ((.status == "COMPLETED")
+            and ((.conclusion // "") | test("^(SUCCESS|NEUTRAL|SKIPPED)$"))) end;
+    def pending:
+      if has("state") then (.state == "PENDING")
+      else (.status != "COMPLETED") end;
+
+    . as $pr
+    | (($pr.statusCheckRollup) // []) as $checks
+    | if $pr.isDraft then "skip: draft"
+      elif ([$pr.labels[]?.name] | any(. as $l | $hold | index($l) != null))
+        then "skip: hold label"
+      elif ($checks | length) == 0 then "skip: no checks reported yet"
+      elif ($checks | map(pending) | any) then "skip: checks still running"
+      elif (($checks | map(ok) | all) | not) then "skip: checks not green"
+      else "merge" end
+  ')
+
+  if [ "$verdict" != "merge" ]; then
+    echo "[auto-merge] #${number} ${verdict} — ${title}"
+
+    # A CANCELLED check is not a verdict, it is noise: CI workflows in this
+    # fleet use `concurrency: cancel-in-progress`, so an unrelated newer run on
+    # the same ref can kill a PR's build. Nothing ever re-runs it, the PR is
+    # never green, and it would sit in this queue forever. Re-run it and let a
+    # later sweep judge the real result. Genuine failures are left alone; only a
+    # run with no real failure is retried.
+    if [ "$verdict" = "skip: checks not green" ]; then
+      retry_urls=$(printf '%s' "$pr" | jq -r '
+        [ .statusCheckRollup[]?
+          | select(has("state") | not)
+          | select((.conclusion // "") == "CANCELLED")
+          | .detailsUrl ] as $cancelled
+        | [ .statusCheckRollup[]?
+            | select(((.conclusion // .state // "")
+                      | test("^(FAILURE|TIMED_OUT|ACTION_REQUIRED|STARTUP_FAILURE|ERROR)$"))) ] as $failed
+        | if ($failed | length) == 0 then $cancelled[] else empty end
+      ')
+      for url in $retry_urls; do
+        run_id=$(printf '%s' "$url" | grep -oE '/runs/[0-9]+' | grep -oE '[0-9]+' || true)
+        [ -z "$run_id" ] && continue
+        echo "[auto-merge] #${number} re-running cancelled run ${run_id}"
+        gh run rerun "$run_id" --repo "$REPO" || echo "[auto-merge] #${number} could not re-run ${run_id}" >&2
+      done
+    fi
+    continue
+  fi
+
+  # Mergeability is computed lazily by GitHub and is invalidated every time the
+  # base branch moves — so right after a merge (exactly when this workflow runs)
+  # every PR reports UNKNOWN. Poll until GitHub has an answer instead of
+  # treating "not computed yet" as "not mergeable"; otherwise the fast path can
+  # never merge anything and the whole train falls back to the cron.
+  mergeable=""
+  state=""
+  for attempt in 1 2 3 4 5 6; do
+    fresh=$(gh pr view "$number" --repo "$REPO" --json mergeable,mergeStateStatus)
+    mergeable=$(printf '%s' "$fresh" | jq -r '.mergeable')
+    state=$(printf '%s' "$fresh" | jq -r '.mergeStateStatus')
+    [ "$mergeable" != "UNKNOWN" ] && break
+    echo "[auto-merge] #${number} mergeability not computed yet (attempt ${attempt}) — waiting"
+    sleep 5
+  done
+
+  if [ "$mergeable" != "MERGEABLE" ]; then
+    echo "[auto-merge] #${number} skip: not mergeable (${mergeable}/${state}) — ${title}"
+    continue
+  fi
+
+  echo "[auto-merge] #${number} green and ready — merging: ${title}"
+  if gh pr merge "$number" --repo "$REPO" --squash --delete-branch; then
+    merged_any=1
+    echo "[auto-merge] #${number} merged"
+    # One car per sweep: let CI verify this on the base before the next couples.
+    break
+  else
+    # Losing a race (someone merged first, or the base moved underneath) is
+    # normal; the next sweep re-evaluates from fresh state.
+    echo "[auto-merge] #${number} merge failed — leaving for the next sweep" >&2
+  fi
+done
+
+if [ "$merged_any" -eq 1 ]; then
+  for wf in $REARM_WORKFLOWS; do
+    echo "[auto-merge] re-arming ${wf} on ${BASE_BRANCH}"
+    gh workflow run "$wf" --repo "$REPO" --ref "$BASE_BRANCH" \
+      || echo "[auto-merge] could not dispatch ${wf} — is workflow_dispatch declared?" >&2
+  done
+else
+  echo "[auto-merge] nothing merged; no re-arm needed"
+fi