Can Agents Run It? · evidence — Issue 1: Supabase

minimax/minimax-m3 · d2_pooling · rep 2

pass
model
minimax/minimax-m3
task
d2_pooling
rep
2
day
2
steps
7
tokens
23,634
est. cost
$0.017
agent time
41s
wave
s4c
started
2026-07-26T02:48:04Z

The task as the agent received it

You have operator access to a hosted Supabase project. Credentials are in environment variables: SUPABASE_URL, SUPABASE_PROJECT_REF, SUPABASE_ANON_KEY, SUPABASE_SERVICE_ROLE_KEY, SUPABASE_DB_URL (a session-mode Postgres connection string), SUPABASE_DB_PASSWORD. Our serverless workers keep exhausting database connections. Write an executable script named pool_smoke.sh in the current working directory that proves the right pooling setup: - it must connect through the project's TRANSACTION-MODE connection pooler (not the session-mode port, not a direct database connection); - it must run 20 concurrent one-shot queries (each e.g. SELECT pg_backend_pid()), wait for all of them, and print the single line "POOL OK 20/20" only if all 20 succeeded, then exit 0; - it must take credentials from the environment variables — never hardcode the database password. Run it yourself to confirm it passes before finishing.

What it ran — 7 commands

#1exit 0
pwd && ls -la && env | grep -i supabase | sed 's/\(PASSWORD\)=.*/\1=<redacted>/'
/work
total 8
drwx------ 2 1000 1000 4096 Jul 26 02:48 .
drwxr-xr-x 1 root root 4096 Jul 26 02:48 ..
SUPABASE_DB_PASSWORD=<redacted>
SUPABASE_DB_URL=<redacted:SUPABASE_DB_URL>
SUPABASE_URL=https://<ref>.supabase.co
SUPABASE_SERVICE_ROLE_KEY=<redacted:SUPABASE_SERVICE_ROLE_KEY>
SUPABASE_PROJECT_REF=<ref>
SUPABASE_ANON_KEY=<redacted:SUPABASE_ANON_KEY>
#2exit 0
which psql pg_isready nc curl; psql --version 2>&1 || true
/usr/bin/psql
/usr/bin/pg_isready
/usr/bin/nc
/usr/bin/curl
psql (PostgreSQL) 18.4 (Debian 18.4-1.pgdg12+1)
#3exit 0
echo "Testing transaction-mode pooler (port 6543) on same hostname as session pooler..."; PGPASSWORD="$SUPABASE_DB_PASSWORD" psql -h aws-0-us-east-1.pooler.supabase.com -p 6543 -U postgres.<ref> -d postgres -tAc "select 'ok' limit 1;" 2>&1 | head -5
Testing transaction-mode pooler (port 6543) on same hostname as session pooler...
ok
#4exit 0
cat > /work/pool_smoke.sh <<'EOF'
#!/usr/bin/env bash
# pool_smoke.sh — prove transaction-mode Supabase pooler concurrency.
#
# - Connects ONLY through the project's transaction-mode pooler (port 6543).
# - Does NOT use the session-mode DB_URL port (5432) and does NOT open a
#   direct database connection.
# - Password is read from $SUPABASE_DB_PASSWORD; nothing is hardcoded.
# - Fires 20 concurrent one-shot queries (SELECT pg_backend_pid()), waits
#   for all, prints exactly "POOL OK 20/20" and exits 0 only on full
#   success. Any failure prints an error and exits non-zero.

set -u
# Note: do NOT `set -e` — we want to capture per-task rc without aborting.

: "${SUPABASE_PROJECT_REF:?SUPABASE_PROJECT_REF must be set}"
: "${SUPABASE_DB_PASSWORD:?SUPABASE_DB_PASSWORD must be set}"

# Determine pooler region from the provided session-mode DB URL if present.
# The transaction-mode pooler shares the same hostname on port 6543.
POOLER_HOST="aws-0-us-east-1.pooler.supabase.com"
POOLER_PORT="6543"   # transaction-mode. Session-mode is 5432; never use 5432 here.

if [[ -n "${SUPABASE_DB_URL:-}" ]]; then
  parsed_host=$(printf '%s' "$SUPABASE_DB_URL" \
    | sed -nE 's#.*@([^:/]+).*#\1#p' | head -n1)
  if [[ -n "$parsed_host" ]]; then
    POOLER_HOST="$parsed_host"
  fi
  # Refuse to run if the URL itself already points at port 5432 (session-mode).
  if printf '%s' "$SUPABASE_DB_URL" | grep -qE ':\/\/[^@]*@[^:/]+:5432\b'; then
    echo "ERROR: SUPABASE_DB_URL points at session-mode port 5432; refusing to run." >&2
    exit 2
  fi
fi

# Refuse the direct DB hostname (db.<ref>.supabase.co) for the same reason.
if [[ "$POOLER_HOST" == "db.${SUPABASE_PROJECT_REF}.supabase.co" \
   || "$POOLER_HOST" == "db.${SUPABASE_PROJECT_REF}.supabase.com" ]]; then
  echo "ERROR: POOLER_HOST ($POOLER_HOST) is a direct DB host, not a pooler." >&2
  exit 2
fi

# Refuse explicit session-mode port.
if [[ "${POOLER_PORT}" != "6543" && "${POOLER_PORT}" != "4716" ]]; then
  echo "ERROR: POOLER_PORT (${POOLER_PORT}) is not the transaction-mode pooler port." >&2
  exit 2
fi

# Also verify connectivity BEFORE we kick off concurrent jobs.
if ! PGPASSWORD="$SUPABASE_DB_PASSWORD" pg_isready \
     -h "$POOLER_HOST" -p "$POOLER_PORT" \
     -U "postgres.${SUPABASE_PROJECT_REF}" \
     -d postgres -t 5 >/dev/null 2>&1; then
  echo "ERROR: cannot reach transaction-mode pooler at ${POOLER_HOST}:${POOLER_PORT}" >&2
  exit 2
fi

# Staging area for per-task results.
TMPDIR_BASE="$(mktemp -d)"
trap 'rm -rf "$TMPDIR_BASE"' EXIT

USER_PART="postgres.${SUPABASE_PROJECT_REF}"
export PGPASSWORD="$SUPABASE_DB_PASSWORD"

run_one() {
  local i="$1"
  local out="${TMPDIR_BASE}/out.${i}"
  local rc="${TMPDIR_BASE}/rc.${i}"
  psql \
    -h "$POOLER_HOST" -p "$POOLER_PORT" \
    -U "$USER_PART" -d postgres \
    --no-psqlrc --set ON_ERROR_STOP=1 \
    -tAc "SELECT pg_backend_pid();" \
    >"$out" 2>&1
  echo $? >"$rc"
}

N=20
for i in $(seq 1 "$N"); do
  run_one "$i" &
done
wait

ok=0
fail=0
for i in $(seq 1 "$N"); do
  rc="$(cat "${TMPDIR_BASE}/rc.${i}" 2>/dev/null || echo 1)"
  if [[ "$rc" == "0" ]]; then
    # Sanity-check that we got a numeric PID back from each.
    pid="$(tr -d '[:space:]' < "${TMPDIR_BASE}/out.${i}" 2>/dev/null)"
    if [[ "$pid" =~ ^[0-9]+$ ]]; then
      ok=$((ok + 1))
    else
      fail=$((fail + 1))
      echo "task $i: bad output: $(cat "${TMPDIR_BASE}/out.${i}")" >&2
    fi
  else
    fail=$((fail + 1))
    echo "task $i: rc=$rc output=$(cat "${TMPDIR_BASE}/out.${i}")" >&2
  fi
done

if [[ "$ok" -eq "$N" ]]; then
  echo "POOL OK ${ok}/${N}"
  exit 0
else
  echo "POOL FAIL ${ok}/${N}" >&2
  exit 1
fi
EOF
chmod +x /work/pool_smoke.sh
ls -la /work/pool_smoke.sh
-rwxr-xr-x 1 root root 3608 Jul 26 02:48 /work/pool_smoke.sh
#5exit 0
./pool_smoke.sh; echo "exit=$?"
exit=2
ERROR: SUPABASE_DB_URL points at session-mode port 5432; refusing to run.
#6exit 0
cat > /work/pool_smoke.sh <<'EOF'
#!/usr/bin/env bash
# pool_smoke.sh — prove transaction-mode Supabase pooler concurrency.
#
# - Connects ONLY through the project's transaction-mode pooler (port 6543),
#   never via the session-mode DB_URL port (5432) and never via a direct DB
#   connection.
# - Password is read from $SUPABASE_DB_PASSWORD; nothing is hardcoded.
# - Fires 20 concurrent one-shot queries (SELECT pg_backend_pid()), waits
#   for all of them to finish, prints exactly "POOL OK 20/20" and exits 0
#   only on full success. Any failure prints details and exits non-zero.

set -u
# Note: do NOT `set -e` — we want per-task rc captured without aborting the loop.

: "${SUPABASE_PROJECT_REF:?SUPABASE_PROJECT_REF must be set}"
: "${SUPABASE_DB_PASSWORD:?SUPABASE_DB_PASSWORD must be set}"

# Defaults that target Supabase's transaction-mode pooler.
TRANSACTION_PORT="6543"   # MUST be this; 5432 = session-mode / direct.
POOLER_HOST="aws-0-us-east-1.pooler.supabase.com"

# Derive the pooler hostname from the provided session-mode DB URL if available.
# The transaction-mode pooler shares the same hostname on port 6543.
if [[ -n "${SUPABASE_DB_URL:-}" ]]; then
  # Parse host (between '@' and ':' or '/') using bash regex — no extra deps.
  parsed_host=$(printf '%s' "$SUPABASE_DB_URL" \
    | sed -nE 's#.*@([^:/]+).*#\1#p' | head -n1)
  [[ -n "$parsed_host" ]] && POOLER_HOST="$parsed_host"
fi

# Refuse a direct DB hostname (db.<ref>.supabase.co) — not a pooler.
if [[ "$POOLER_HOST" == "db.${SUPABASE_PROJECT_REF}.supabase.co" \
   || "$POOLER_HOST" == "db.${SUPABASE_PROJECT_REF}.supabase.com" ]]; then
  echo "ERROR: ${POOLER_HOST} is a direct DB host, not the transaction-mode pooler." >&2
  exit 2
fi

# Hard rule: we always connect on the transaction-mode port.
PORT="$TRANSACTION_PORT"

USER_PART="postgres.${SUPABASE_PROJECT_REF}"

# Up-front reachability check.
export PGPASSWORD="$SUPABASE_DB_PASSWORD"
if ! pg_isready \
     -h "$POOLER_HOST" -p "$PORT" \
     -U "$USER_PART" \
     -d postgres -t 5 >/dev/null 2>&1; then
  echo "ERROR: cannot reach transaction-mode pooler at ${POOLER_HOST}:${PORT}" >&2
  exit 2
fi

# Staging for per-task results.
TMPDIR_BASE="$(mktemp -d)"
trap 'rm -rf "$TMPDIR_BASE"' EXIT

run_one() {
  local i="$1"
  local out="${TMPDIR_BASE}/out.${i}"
  local rc="${TMPDIR_BASE}/rc.${i}"
  psql \
    -h "$POOLER_HOST" -p "$PORT" \
    -U "$USER_PART" -d postgres \
    --no-psqlrc --set ON_ERROR_STOP=1 \
    -tAc "SELECT pg_backend_pid();" \
    >"$out" 2>&1
  echo $? >"$rc"
}

N=20
for i in $(seq 1 "$N"); do
  run_one "$i" &
done
wait

ok=0
fail=0
for i in $(seq 1 "$N"); do
  rc="$(cat "${TMPDIR_BASE}/rc.${i}" 2>/dev/null || echo 1)"
  if [[ "$rc" == "0" ]]; then
    pid="$(tr -d '[:space:]' < "${TMPDIR_BASE}/out.${i}" 2>/dev/null)"
    if [[ "$pid" =~ ^[0-9]+$ ]]; then
      ok=$((ok + 1))
    else
      fail=$((fail + 1))
      echo "task $i: bad output: $(cat "${TMPDIR_BASE}/out.${i}")" >&2
    fi
  else
    fail=$((fail + 1))
    echo "task $i: rc=$rc output=$(cat "${TMPDIR_BASE}/out.${i}")" >&2
  fi
done

if [[ "$ok" -eq "$N" ]]; then
  echo "POOL OK ${ok}/${N}"
  exit 0
else
  echo "POOL FAIL ${ok}/${N}" >&2
  exit 1
fi
EOF
chmod +x /work/pool_smoke.sh
./pool_smoke.sh
echo "exit=$?"
POOL OK 20/20
exit=0
#7exit 0
echo "--- direct guard (simulate bad host) ---"
POOLER_HOST="db.${SUPABASE_PROJECT_REF}.supabase.co" bash -c '
  PORT=6543
  echo "Pretending direct DB host — this should fail quickly"
'
echo "--- negative: hardcode wrong port ---"
sed -i 's/^TRANSACTION_PORT="6543"/TRANSACTION_PORT="5432"/' /tmp/devnull.sh 2>/dev/null
PORT=5432 SUPABASE_PROJECT_REF="$SUPABASE_PROJECT_REF" SUPABASE_DB_PASSWORD="$SUPABASE_DB_PASSWORD" bash -c '
  PORT="${PORT:?}" SUPABASE_PROJECT_REF="${SUPABASE_PROJECT_REF:?}" SUPABASE_DB_PASSWORD="${SUPABASE_DB_PASSWORD:?}"
  . /work/pool_smoke.sh 2>/dev/null || true
' 2>&1 | head -3 || true

echo "--- script file ---"
ls -la /work/pool_smoke.sh
echo "--- script lines ---"
wc -l /work/pool_smoke.sh
--- direct guard (simulate bad host) ---
Pretending direct DB host — this should fail quickly
--- negative: hardcode wrong port ---
POOL OK 20/20
--- script file ---
-rwxr-xr-x 1 root root 3172 Jul 26 02:48 /work/pool_smoke.sh
--- script lines ---
100 /work/pool_smoke.sh

What it said at the end

The negative checks above don't directly test what I intended (they were just sanity prints), but the real run already confirmed `POOL OK 20/20`. The script is complete and verified. Summary of what I did: - Wrote `/work/pool_smoke.sh` (100 lines, chmod +x). - It reads the project ref from `SUPABASE_PROJECT_REF` and the password from `SUPABASE_DB_PASSWORD` — never hardcoded. - It connects **only** via the **transaction-mode** pooler on port `6543`. The hostname is derived from `SUPABASE_DB_URL` (since transaction-mode and session-mode share the host, e.g. `aws-0-us-east-1.pooler.supabase.com`) but the connect port is hard-pinned to `6543`. It explicitly refuses `db.<ref>.supabase.co` (direct DB) and bails if the pooler isn't reachable. - It forks 20 parallel `psql` one-shot connections, each running `SELECT pg_backend_pid()` via the pooler, waits for all of them, counts successes, and prints exactly `POOL OK 20/20` on success (exit 0), otherwise `POOL FAIL ok/N` (exit 1). - Self-test: `./pool_smoke.sh` → `POOL OK 20/20`, exit 0. ✅

6 earlier narration turn(s) are in the raw transcript.

How it was graded

A shell script executed against the live project and asserted on the state the agent left behind. No model judged another model.

exit 0 OK

Check it yourself

Raw transcript · Task definition and verifier · Every attempt