Add suspicious-signup sweep: suspend flagged signups with no activity

Watches every account flagged by IP-scrutiny or email-domain scrutiny at
signup and, once it goes live, records a baseline post/follow count. A
scheduled sweep (app/suspicious_sweep.py, run via cron on admin.yttrx.com)
suspends any watch past SUSPICIOUS_GRACE_HOURS with zero new posts and
zero new follows since that baseline; any activity clears the watch.

Works whether yttrx is open-registration or requires moderator approval,
since the watch starts at whichever event actually makes the account live
(account.created vs account.approved), same dual handling the welcome flow
already uses. Needs ABUSE_BOT_TOKEN re-minted with admin:read:accounts.

.env.example

@@ -145,3 +145,35 @@ CHECK_MAIL_HOLD_WELCOME=true
 # 403s and is logged as an error, but nothing else in the bot is affected.
 # Also gates the report-triggered suspend+block override in handle_report().
 CHECK_MAIL_AUTO_DOMAIN_BLOCK=true
+
+# --- Suspicious-signup sweep (app/suspicious_sweep.py, run on a schedule) ---
+# Watches every account flagged by EITHER IP-scrutiny or email-domain
+# scrutiny at signup. Once the account goes live (account.created if open/
+# auto-approved, account.approved if this instance requires moderator
+# approval — same dual handling the welcome flow already uses), a baseline
+# post/follow count is recorded. A scheduled sweep (cron on admin.yttrx.com,
+# `docker exec yttrx-welcomebot python -m app.suspicious_sweep`) then acts on
+# any watch past SUSPICIOUS_GRACE_HOURS with zero new posts AND zero new
+# follows since that baseline. Master switch:
+SUSPICIOUS_SWEEP_ENABLED=true
+
+# Hours of grace given to a flagged signup to show organic activity (a new
+# post or a new follow) before it's considered dormant/bot-like.
+SUSPICIOUS_GRACE_HOURS=24
+
+# Moderation action taken on a flagged signup with no activity in the grace
+# window: "suspend" (agreed default — a stronger signal than the report-
+# driven abuse-bot's default "silence", since this is zero organic activity,
+# not just reports) or "silence".
+SUSPICIOUS_ACTION=suspend
+
+# Rollout safety switch, same role as ABUSE_DRY_RUN — ships "false" (live)
+# here by design; flip to "true" to pause without redeploying.
+SUSPICIOUS_DRY_RUN=false
+
+# Requires ABUSE_BOT_TOKEN to additionally carry the admin:read:accounts
+# scope (fetches current post/follow counts + role/suspended state via
+# GET /api/v1/admin/accounts/:id) — see CLAUDE.md's moderator-token-gotcha
+# section. Without it the sweep 403s per-account (logged as an error, that
+# account is retried next sweep) but nothing else in the bot is affected.
+# Reuses ABUSE_SKIP_PRIVILEGED and ABUSE_ALLOWLIST above — no separate vars.

CLAUDE.md

@@ -12,11 +12,18 @@ A FastAPI webhook server. One Mastodon admin webhook delivers `account.created`,
   (datacenter/hosting) signups get more scrutiny (held welcome, ip_blocks
   registration, DM to moderator) before falling through to a normal welcome.
 - **account.approved** → always DM the welcome (dedup-safe; this is what
-  releases a held welcome once a human clears the approval queue).
+  releases a held welcome once a human clears the approval queue). Also where
+  a held, flagged signup's suspicious-watch baseline gets captured.
 - **report.created** → classify the reported account, count distinct reporters,
   and auto-**silence** young/dormant accounts past the threshold — lowered for
   accounts with a flagged signup IP (see README).
 
+A separate scheduled module, `app/suspicious_sweep.py` (run via
+`python -m app.suspicious_sweep`, not part of the webhook server), suspends
+any IP- or email-flagged signup that shows zero new posts and zero new
+follows within `SUSPICIOUS_GRACE_HOURS` of going live. See README.md's
+"Suspicious-signup sweep" section for the full policy.
+
 | Where | What |
 |---|---|
 | Workstation | Source of truth: `~/yttrx-welcomebot/` (this repo) |
@@ -170,12 +177,18 @@ as an error; nothing else in the bot is affected).
 report-triggered suspend override) will 403 (logged as an error; the suspend
 itself still goes through since that only needs `admin:write:accounts`).
 
+**The suspicious-signup sweep (`app/suspicious_sweep.py`) needs another scope,
+`admin:read:accounts`**, to `GET /api/v1/admin/accounts/:id` (fetches current
+post/follow counts + role/suspended state to decide whether to suspend) —
+without it every sweep run 403s per-account (logged as an error, that account
+is simply retried next run; nothing else in the bot is affected).
+
 Mint a correctly-scoped token from the Rails console on mammut:
 
 ```bash
 ssh mammut 'docker exec $(docker ps -f name=live-web-1 -q) bin/rails runner "
 acct = Account.find_local(%q{bot})
-scopes = %q{read:statuses write:statuses admin:read:reports admin:write:accounts admin:write:ip_blocks admin:write:email_domain_blocks}
+scopes = %q{read:statuses write:statuses admin:read:reports admin:write:accounts admin:write:ip_blocks admin:write:email_domain_blocks admin:read:accounts}
 app = Doorkeeper::Application.create!(name: %q{Abuse handler vN}, scopes: scopes, redirect_uri: %q{urn:ietf:wg:oauth:2.0:oob})
 tok = Doorkeeper::AccessToken.create!(application_id: app.id, resource_owner_id: acct.user.id, scopes: scopes)
 puts tok.token
@@ -213,6 +226,14 @@ curl -s -o /dev/null -w "%{http_code}\n" -X POST \
   https://yttrx.com/api/v1/admin/email_domain_blocks
 ```
 
+And for `admin:read:accounts` (`403` = missing scope; `200`/`404` = authorized):
+
+```bash
+curl -s -o /dev/null -w "%{http_code}\n" \
+  -H "Authorization: Bearer $ABUSE_BOT_TOKEN" \
+  https://yttrx.com/api/v1/admin/accounts/0
+```
+
 ## Mastodon side (mammut)
 
 - The webhook (Administration → Webhooks) is already subscribed to

README.md

@@ -124,14 +124,57 @@ Registering email_domain_blocks requires the `ABUSE_BOT_TOKEN` to carry the
 `admin:write:email_domain_blocks` scope in addition to its existing scopes
 (see `CLAUDE.md`).
 
+## Suspicious-signup sweep
+
+A scheduled companion to the two signup-scrutiny signals above (`app/suspicious_sweep.py`,
+run via `docker exec yttrx-welcomebot python -m app.suspicious_sweep`, e.g. hourly cron on
+`admin.yttrx.com`):
+
+1. Any account flagged by **either** IP-scrutiny (datacenter/hosting) or
+   email-domain scrutiny (disposable/high-risk) has a grace-period clock
+   started the moment it goes live — `account.created` if signups are open
+   (or auto-approved), `account.approved` if this instance requires
+   moderator approval. Same dual-event handling the welcome flow already
+   uses, so it works in either registration mode without extra config. A
+   baseline snapshot of `statuses_count`/`following_count` is recorded at
+   that moment (`app.main.maybe_start_suspicious_watch`, table
+   `suspicious_watch`).
+2. The sweep looks for watches past `SUSPICIOUS_GRACE_HOURS` (default 24)
+   and compares current counts to the baseline:
+   - **Any new post or new follow** since the baseline → cleared, never
+     rechecked.
+   - **Zero of either** → `SUSPICIOUS_ACTION` (default `suspend`) is applied,
+     the moderator is DMed, and the watch is marked resolved.
+3. Already-suspended/silenced accounts, staff (`ABUSE_SKIP_PRIVILEGED`), and
+   `ABUSE_ALLOWLIST` handles are skipped exactly as elsewhere in this bot —
+   the sweep reuses those same settings rather than defining its own.
+
+Unlike the report-driven abuse-bot (default `silence`), this sweep defaults
+to **`suspend`** — the reasoning is that zero organic activity in a full day
+is a stronger bulk/bot-signup signal than a report alone, and Mastodon
+suspensions have a server-side undo window. It also ships **live**
+(`SUSPICIOUS_DRY_RUN=false`) rather than with the usual dry-run rollout step;
+`SUSPICIOUS_DRY_RUN=true` remains available as a kill switch if you want to
+pause it without redeploying.
+
+Requires `ABUSE_BOT_TOKEN` to additionally carry the **`admin:read:accounts`**
+scope (fetches live counts + role/suspended state via
+`GET /api/v1/admin/accounts/:id`) — see `CLAUDE.md`'s moderator-token-gotcha
+section for the remint procedure.
+
+Inspect the watch list: `welcomebot-suspicious` / `welcomebot-suspicious --pending`
+(see `bin/welcomebot-suspicious`).
+
 ## Layout
 
 | Path | Purpose |
 |---|---|
 | `app/main.py` | The FastAPI app |
+| `app/suspicious_sweep.py` | Scheduled sweep: suspends flagged signups with no activity (run via cron, not the webserver) |
 | `Dockerfile`, `docker-compose.yml` | Container build + run |
 | `.env.example` | Config template (copy to `.env`, never commit) |
 | `nginx/hooks.yttrx.com.conf` | nginx site for admin.yttrx.com |
+| `bin/welcomebot-logs`, `bin/welcomebot-signups`, `bin/welcomebot-suspicious` | Ops CLIs installed to `/root/bin` on admin.yttrx.com |
 | `test_local.py` | Offline smoke test (no network) |
 
 ## Configuration
@@ -172,6 +215,10 @@ Copy `.env.example` to `.env` and fill in:
 | `CHECK_MAIL_RISK_THRESHOLD` | Flag if `is_disposable` or `risk` ≥ this (default 80) |
 | `CHECK_MAIL_HOLD_WELCOME` | `true` — hold the welcome for a flagged signup until `account.approved` |
 | `CHECK_MAIL_AUTO_DOMAIN_BLOCK` | Auto-register a flagged domain into Mastodon's `Admin::EmailDomainBlock`; also gates the report-triggered override |
+| `SUSPICIOUS_SWEEP_ENABLED` | Master switch for the suspicious-signup sweep (`app/suspicious_sweep.py`) |
+| `SUSPICIOUS_GRACE_HOURS` | Hours a flagged signup has to post or follow someone before it's swept (default 24) |
+| `SUSPICIOUS_ACTION` | `suspend` (default) or `silence`, applied to a flagged signup with zero activity in the grace window |
+| `SUSPICIOUS_DRY_RUN` | `false` (ships live by design) — set `true` to log/DM only, no real action |
 
 ## Local test
 
@@ -272,11 +319,29 @@ Register a throwaway test account on yttrx and confirm the `@welcome` bot DMs
 it. Then delete the test account (`tootctl accounts delete` on mammut) and the
 test toot.
 
+### 6. Cron for the suspicious-signup sweep (on admin.yttrx.com)
+
+The sweep is not part of the always-running webhook server — it's invoked
+on a schedule via `docker exec`. `ABUSE_BOT_TOKEN` must additionally carry
+the `admin:read:accounts` scope (remint via the Rails-console snippet in
+`CLAUDE.md`, same procedure as the other scope additions) before installing
+this cron entry, or every sweep run 403s per-account:
+
+```bash
+crontab -l | { cat; echo '17 * * * * docker exec yttrx-welcomebot python -m app.suspicious_sweep >> /var/log/welcomebot-suspicious-sweep.log 2>&1'; } | crontab -
+```
+
+Hourly (rather than daily like the dormant-sweep) since the grace window is
+only `SUSPICIOUS_GRACE_HOURS` (default 24) — an hourly cadence keeps the gap
+between crossing the deadline and being swept small.
+
 ## Operations
 
 ```bash
 welcomebot-logs -f --abuse             # convenience CLI on admin (see bin/welcomebot-logs)
 welcomebot-signups --flagged           # signup-IP classification history (see bin/welcomebot-signups)
+welcomebot-suspicious --pending        # suspicious-sweep watch list (see bin/welcomebot-suspicious)
+docker exec yttrx-welcomebot python -m app.suspicious_sweep   # run the sweep manually
 docker compose logs -f welcomebot      # watch deliveries
 docker compose restart welcomebot      # after editing .env
 docker compose down && docker compose up -d --build   # redeploy after code change

app/main.py

@@ -170,6 +170,29 @@ CHECK_MAIL_HOLD_WELCOME = os.environ.get("CHECK_MAIL_HOLD_WELCOME", "true").lowe
 # Admin::EmailDomainBlock (blocks future signups from that domain).
 CHECK_MAIL_AUTO_DOMAIN_BLOCK = os.environ.get("CHECK_MAIL_AUTO_DOMAIN_BLOCK", "true").lower() in ("1", "true", "yes")
 
+# --- Suspicious-signup sweep (flagged at signup, no activity in grace window) ---
+# Watches every account flagged by EITHER IP-scrutiny or email-domain scrutiny
+# at signup. Once the account goes live — account.created if signups are open
+# (or an already-approved account), account.approved if this instance requires
+# moderator approval, same dual event handling the welcome flow already uses —
+# a baseline post/follow count is recorded (see maybe_start_suspicious_watch).
+# A separate scheduled job (bin/welcomebot-suspicious sweep, see CLAUDE.md) run
+# via `python -m app.suspicious_sweep` looks for accounts past
+# SUSPICIOUS_GRACE_HOURS with zero new posts AND zero new follows since that
+# baseline, and takes SUSPICIOUS_ACTION on them.
+SUSPICIOUS_SWEEP_ENABLED = os.environ.get("SUSPICIOUS_SWEEP_ENABLED", "true").lower() in ("1", "true", "yes")
+# Hours of grace given to a flagged signup to show organic activity (a new
+# post or a new follow) before it's considered dormant/bot-like.
+SUSPICIOUS_GRACE_HOURS = int(os.environ.get("SUSPICIOUS_GRACE_HOURS", "24"))
+# "suspend" (agreed default — a flagged signup with zero organic activity in
+# the grace window is a strong bulk/bot-signup signal, stronger than the
+# report-driven abuse-bot's default "silence") or "silence".
+SUSPICIOUS_ACTION = os.environ.get("SUSPICIOUS_ACTION", "suspend").lower()
+# Rollout safety switch, same role as ABUSE_DRY_RUN — ships "false" (live)
+# here per an explicit decision to skip the usual dry-run rollout step for
+# this feature. Flip to "true" to pause without redeploying.
+SUSPICIOUS_DRY_RUN = os.environ.get("SUSPICIOUS_DRY_RUN", "false").lower() in ("1", "true", "yes")
+
 if not WEBHOOK_SECRET:
     log.warning("WEBHOOK_SECRET is empty — signature verification will reject all requests.")
 if not BOT_ACCESS_TOKEN:
@@ -248,6 +271,18 @@ def _init_db() -> None:
             "  cached_at  TEXT DEFAULT CURRENT_TIMESTAMP"
             ")"
         )
+        conn.execute(
+            "CREATE TABLE IF NOT EXISTS suspicious_watch ("
+            "  account_id TEXT PRIMARY KEY,"
+            "  acct       TEXT,"
+            "  reasons    TEXT,"
+            "  started_at TEXT DEFAULT CURRENT_TIMESTAMP,"
+            "  baseline_statuses_count  INTEGER,"
+            "  baseline_following_count INTEGER,"
+            "  status     TEXT DEFAULT 'pending',"
+            "  resolved_at TEXT"
+            ")"
+        )
 
 
 @contextmanager
@@ -387,6 +422,25 @@ def cache_check_mail(domain: str, is_disposable: bool, risk: int) -> None:
         )
 
 
+def already_watched_suspicious(account_id: str) -> bool:
+    with _db() as conn:
+        row = conn.execute(
+            "SELECT 1 FROM suspicious_watch WHERE account_id = ?", (account_id,)
+        ).fetchone()
+        return row is not None
+
+
+def start_suspicious_watch(account_id: str, acct: str, reasons: str,
+                           statuses_count: int, following_count: int) -> None:
+    with _db() as conn:
+        conn.execute(
+            "INSERT OR IGNORE INTO suspicious_watch "
+            "(account_id, acct, reasons, baseline_statuses_count, baseline_following_count) "
+            "VALUES (?, ?, ?, ?, ?)",
+            (account_id, acct, reasons, statuses_count, following_count),
+        )
+
+
 _init_db()
 
 # --- Signature verification ------------------------------------------------
@@ -542,7 +596,51 @@ def process_signup(account_id: str, acct: str, ip: str, email: str = "") -> None
 
     if not hold:
         send_welcome(account_id, acct)
-    # else: held — sent later if/when account.approved fires (human review).
+        maybe_start_suspicious_watch(account_id, acct)
+    # else: held — sent later if/when account.approved fires (human review),
+    # which is also where maybe_start_suspicious_watch gets called.
+
+
+def maybe_start_suspicious_watch(account_id: str, acct: str) -> None:
+    """Start the suspicious-sweep grace-period clock for a flagged signup.
+
+    Called at the moment the account actually goes live — immediately in
+    process_signup() when the welcome isn't held, or from the
+    account.approved handler once a human clears the approval queue for a
+    held one. Either path is a no-op if the account was never flagged (by
+    IP-scrutiny or email-domain scrutiny) or is already being watched, so
+    it's safe to call from both places unconditionally.
+    """
+    if not SUSPICIOUS_SWEEP_ENABLED or already_watched_suspicious(account_id):
+        return
+
+    ip_flagged = get_signup_flag(account_id)
+    _, email_flagged = get_signup_email(account_id)
+    if not (ip_flagged or email_flagged):
+        return
+
+    reasons = ",".join(
+        r for r, flagged in (("ip", ip_flagged), ("email", email_flagged)) if flagged
+    )
+    try:
+        statuses_count, following_count = fetch_account_counts(account_id)
+    except httpx.HTTPError as exc:
+        log.error("failed to snapshot suspicious-watch baseline for acct=%s: %s", acct, exc)
+        return
+
+    start_suspicious_watch(account_id, acct, reasons, statuses_count, following_count)
+    log.info("started suspicious watch acct=%s reasons=%s baseline=(statuses=%d, following=%d)",
+             acct, reasons, statuses_count, following_count)
+
+
+def fetch_account_counts(account_id: str) -> tuple[int, int]:
+    """Return (statuses_count, following_count) for an account via the public
+    account API. Split out from maybe_start_suspicious_watch as a single seam
+    to stub in tests."""
+    resp = httpx.get(f"{MASTODON_BASE_URL}/api/v1/accounts/{account_id}", timeout=15.0)
+    resp.raise_for_status()
+    data = resp.json()
+    return data.get("statuses_count", 0), data.get("following_count", 0)
 
 
 # --- Mastodon API (abuse) --------------------------------------------------
@@ -945,8 +1043,13 @@ def _handle_account_created(obj: dict, background: BackgroundTasks, event: str):
     else:
         # account.approved: a human has cleared the approval-required
         # registration gate. Always welcome (dedup makes this safe) — this is
-        # what lifts a held welcome for a previously-flagged signup.
+        # what lifts a held welcome for a previously-flagged signup. Also the
+        # moment a held, flagged signup's suspicious-watch clock should start;
+        # a no-op if it was never flagged or already started (open-signup
+        # case, where account.approved may fire right alongside
+        # account.created).
         background.add_task(send_welcome, account_id, acct)
+        background.add_task(maybe_start_suspicious_watch, account_id, acct)
     return {"ok": True, "queued": acct}
 
 

app/suspicious_sweep.py

@@ -0,0 +1,167 @@
+"""Suspicious-signup sweep.
+
+Suspends (SUSPICIOUS_ACTION) accounts that were flagged at signup by either
+IP-scrutiny (datacenter/hosting IP) or email-domain scrutiny (disposable/
+high-risk domain) and show zero new posts AND zero new follows within
+SUSPICIOUS_GRACE_HOURS of going live.
+
+Only acts on accounts app.main.maybe_start_suspicious_watch already recorded
+in the `suspicious_watch` table — this module doesn't touch signup
+classification itself, just resolves the grace-period clock those signals
+started.
+
+Run on a schedule (cron on admin.yttrx.com, see CLAUDE.md):
+
+    docker exec yttrx-welcomebot python -m app.suspicious_sweep
+
+Requires ABUSE_BOT_TOKEN to carry the admin:read:accounts scope in addition
+to its existing scopes (see CLAUDE.md's moderator-token-gotcha section) —
+without it, fetching current account state 403s and that account is skipped
+(logged as an error) until the token is re-minted.
+"""
+
+from __future__ import annotations
+
+import logging
+from datetime import datetime, timedelta, timezone
+
+import httpx
+
+from app.main import (
+    ABUSE_ALLOWLIST,
+    MASTODON_BASE_URL,
+    SUSPICIOUS_ACTION,
+    SUSPICIOUS_DRY_RUN,
+    SUSPICIOUS_GRACE_HOURS,
+    SUSPICIOUS_SWEEP_ENABLED,
+    ABUSE_SKIP_PRIVILEGED,
+    _admin_headers,
+    _db,
+    apply_action,
+    dm_moderator,
+)
+
+logging.basicConfig(
+    level=logging.INFO,
+    format="%(asctime)s %(levelname)s %(name)s %(message)s",
+)
+log = logging.getLogger("welcomebot.suspicious_sweep")
+
+
+def _pending_due() -> list[tuple]:
+    cutoff = (datetime.now(timezone.utc) - timedelta(hours=SUSPICIOUS_GRACE_HOURS)).isoformat()
+    with _db() as conn:
+        return conn.execute(
+            "SELECT account_id, acct, reasons, baseline_statuses_count, "
+            "       baseline_following_count "
+            "FROM suspicious_watch WHERE status = 'pending' AND started_at <= ?",
+            (cutoff,),
+        ).fetchall()
+
+
+def _resolve(account_id: str, status: str) -> None:
+    with _db() as conn:
+        conn.execute(
+            "UPDATE suspicious_watch SET status = ?, resolved_at = CURRENT_TIMESTAMP "
+            "WHERE account_id = ?",
+            (status, account_id),
+        )
+
+
+def _fetch_admin_account(account_id: str) -> dict | None:
+    """GET the Admin::Account entity (role, suspended/silenced, nested public
+    account with live statuses_count/following_count). Returns None if the
+    account no longer exists (e.g. deleted since being flagged)."""
+    resp = httpx.get(
+        f"{MASTODON_BASE_URL}/api/v1/admin/accounts/{account_id}",
+        headers=_admin_headers(), timeout=15.0,
+    )
+    if resp.status_code == 404:
+        return None
+    resp.raise_for_status()
+    return resp.json()
+
+
+def _is_privileged(admin_acct: dict) -> bool:
+    role = admin_acct.get("role") or {}
+    try:
+        role_id = int(role.get("id")) if role.get("id") is not None else None
+    except (TypeError, ValueError):
+        role_id = None
+    return role_id is not None and role_id > 0
+
+
+def _is_allowlisted(acct: str) -> bool:
+    return acct.split("@")[0].lower() in ABUSE_ALLOWLIST or acct.lower() in ABUSE_ALLOWLIST
+
+
+def sweep() -> None:
+    if not SUSPICIOUS_SWEEP_ENABLED:
+        log.info("SUSPICIOUS_SWEEP_ENABLED=false, nothing to do")
+        return
+
+    due = _pending_due()
+    log.info("%d flagged signup(s) past the %dh grace window", len(due), SUSPICIOUS_GRACE_HOURS)
+
+    for account_id, acct, reasons, baseline_statuses, baseline_following in due:
+        try:
+            admin_acct = _fetch_admin_account(account_id)
+        except httpx.HTTPError as exc:
+            log.error("acct=%s: failed to fetch admin account, will retry next sweep: %s", acct, exc)
+            continue
+
+        if admin_acct is None:
+            log.info("acct=%s no longer exists, clearing watch", acct)
+            _resolve(account_id, "skipped-not-found")
+            continue
+
+        if admin_acct.get("suspended") or admin_acct.get("silenced"):
+            log.info("acct=%s already limited by another mechanism, clearing watch", acct)
+            _resolve(account_id, "skipped-already-limited")
+            continue
+
+        if (_is_privileged(admin_acct) and ABUSE_SKIP_PRIVILEGED) or _is_allowlisted(acct):
+            log.info("acct=%s is staff/allowlisted, clearing watch", acct)
+            _resolve(account_id, "skipped-privileged")
+            continue
+
+        public = admin_acct.get("account") or {}
+        statuses_count = public.get("statuses_count", 0)
+        following_count = public.get("following_count", 0)
+
+        if statuses_count > baseline_statuses or following_count > baseline_following:
+            log.info(
+                "acct=%s showed activity since flagging (posts %d->%d, following %d->%d), clearing watch",
+                acct, baseline_statuses, statuses_count, baseline_following, following_count,
+            )
+            _resolve(account_id, "cleared")
+            continue
+
+        note = (
+            f"Auto-{SUSPICIOUS_ACTION} by suspicious-sweep: flagged at signup ({reasons}), "
+            f"no post or follow in {SUSPICIOUS_GRACE_HOURS}h since going live."
+        )
+
+        if SUSPICIOUS_DRY_RUN:
+            log.warning("[DRY-RUN] would %s acct=%s (reasons=%s, no activity in %dh)",
+                        SUSPICIOUS_ACTION, acct, reasons, SUSPICIOUS_GRACE_HOURS)
+            _resolve(account_id, f"dry-run:{SUSPICIOUS_ACTION}")
+            dm_moderator(f"[DRY-RUN] would {SUSPICIOUS_ACTION} @{acct} — flagged signup "
+                         f"({reasons}), no activity in {SUSPICIOUS_GRACE_HOURS}h.")
+            continue
+
+        try:
+            apply_action(account_id, SUSPICIOUS_ACTION, note)
+        except httpx.HTTPError as exc:
+            log.error("acct=%s: failed to %s, will retry next sweep: %s", acct, SUSPICIOUS_ACTION, exc)
+            continue
+
+        _resolve(account_id, SUSPICIOUS_ACTION)
+        log.warning("auto-%sd acct=%s — flagged signup (%s), no activity in %dh",
+                    SUSPICIOUS_ACTION, acct, reasons, SUSPICIOUS_GRACE_HOURS)
+        dm_moderator(f"🚨 Auto-{SUSPICIOUS_ACTION}d @{acct} — flagged signup ({reasons}), "
+                     f"no post or follow in {SUSPICIOUS_GRACE_HOURS}h since going live.")
+
+
+if __name__ == "__main__":
+    sweep()

bin/welcomebot-suspicious

@@ -0,0 +1,93 @@
+#!/usr/bin/env bash
+# welcomebot-suspicious — inspect the suspicious-sweep watch list for yttrx-welcomebot.
+#
+# Installed at /usr/local/bin/welcomebot-suspicious on admin.yttrx.com.
+# Source of truth: ~/yttrx-welcomebot/bin/welcomebot-suspicious (workstation).
+# Reads the persistent `suspicious_watch` sqlite table — see app/main.py's
+# maybe_start_suspicious_watch / app/suspicious_sweep.py.
+set -euo pipefail
+
+CONTAINER="${WELCOMEBOT_CONTAINER:-yttrx-welcomebot}"
+limit=20
+pending_only=0
+acct_filter=""
+
+usage() {
+  cat <<EOF
+Usage: welcomebot-suspicious [options]
+
+Print suspicious-watch records (acct, flag reasons, baseline post/follow
+counts, status, timestamps) from the welcomebot's persistent sqlite store,
+newest first.
+
+Options:
+  -n, --limit N     Show the last N rows (default: $limit; 'all' for everything).
+      --pending     Only rows still waiting out the grace period.
+  -a, --acct ACCT   Only rows for this acct (exact match).
+  -h, --help        Show this help.
+
+Examples:
+  welcomebot-suspicious                # last $limit watch records
+  welcomebot-suspicious --pending      # still in the grace window
+  welcomebot-suspicious -n all
+  welcomebot-suspicious -a someuser
+
+Run the sweep itself (normally invoked by cron, see CLAUDE.md):
+  ssh admin.yttrx.com 'docker exec yttrx-welcomebot python -m app.suspicious_sweep'
+EOF
+}
+
+while [ $# -gt 0 ]; do
+  case "$1" in
+    -n|--limit)   shift; limit="${1:?--limit needs a value}" ;;
+    --pending)    pending_only=1 ;;
+    -a|--acct)    shift; acct_filter="${1:?--acct needs a value}" ;;
+    -h|--help)    usage; exit 0 ;;
+    *) echo "welcomebot-suspicious: unknown option: $1" >&2; usage >&2; exit 2 ;;
+  esac
+  shift
+done
+
+if ! command -v docker >/dev/null 2>&1; then
+  echo "welcomebot-suspicious: docker not found in PATH" >&2; exit 1
+fi
+if ! docker inspect "$CONTAINER" >/dev/null 2>&1; then
+  echo "welcomebot-suspicious: container '$CONTAINER' not found (is it deployed/running?)" >&2; exit 1
+fi
+
+docker exec -i \
+  -e SUS_LIMIT="$limit" \
+  -e SUS_PENDING_ONLY="$pending_only" \
+  -e SUS_ACCT="$acct_filter" \
+  "$CONTAINER" python3 <<'PYEOF'
+import os
+import sqlite3
+
+conn = sqlite3.connect("/data/welcomed.db")
+where, params = [], []
+if os.environ.get("SUS_PENDING_ONLY") == "1":
+    where.append("status = 'pending'")
+if os.environ.get("SUS_ACCT"):
+    where.append("acct = ?")
+    params.append(os.environ["SUS_ACCT"])
+
+sql = ("SELECT acct, reasons, baseline_statuses_count, baseline_following_count, "
+       "status, started_at, resolved_at FROM suspicious_watch")
+if where:
+    sql += " WHERE " + " AND ".join(where)
+sql += " ORDER BY started_at DESC"
+
+limit = os.environ.get("SUS_LIMIT", "20")
+if limit != "all":
+    sql += f" LIMIT {int(limit)}"
+
+rows = conn.execute(sql, params).fetchall()
+if not rows:
+    print("(no matching suspicious-watch records)")
+for acct, reasons, base_statuses, base_following, status, started_at, resolved_at in rows:
+    acct = acct or ""
+    reasons = reasons or ""
+    baseline = f"posts={base_statuses} following={base_following}"
+    resolved = resolved_at or "-"
+    print(f"{started_at}  {acct:20s} {reasons:10s} {baseline:24s} {status:20s} resolved={resolved}")
+PYEOF

test_local.py

@@ -225,6 +225,9 @@ def ip_scrutiny_tests():
     main.send_welcome = lambda account_id, acct: sent.append((account_id, acct))
     main.dm_moderator = lambda message: dms.append(message)
     main.register_ip_block = lambda ip, acct, org: ipblocks.append((ip, acct, org))
+    # A flagged-but-not-held signup starts the suspicious watch inline, which
+    # would otherwise hit the network for a baseline snapshot — stub it.
+    main.fetch_account_counts = lambda account_id: (0, 0)
 
     def classify(ip):
         return classifications[ip]
@@ -302,6 +305,9 @@ def email_scrutiny_tests():
     main.send_welcome = lambda account_id, acct: sent.append((account_id, acct))
     main.dm_moderator = lambda message: dms.append(message)
     main.register_email_domain_block = lambda domain, acct: domain_blocks.append((domain, acct))
+    # A flagged-but-not-held signup starts the suspicious watch inline, which
+    # would otherwise hit the network for a baseline snapshot — stub it.
+    main.fetch_account_counts = lambda account_id: (0, 0)
 
     classifications = {
         "gmail.com": (False, 5),
@@ -387,9 +393,155 @@ def email_scrutiny_tests():
     assert domain_blocks == [], domain_blocks
 
 
+def suspicious_watch_tests():
+    """Drive maybe_start_suspicious_watch: baseline capture on flagged
+    signups only, never on clean ones, and never twice for the same account."""
+    main.SUSPICIOUS_SWEEP_ENABLED = True
+    counts = {"301": (2, 5), "302": (0, 0)}
+    main.fetch_account_counts = lambda account_id: counts[account_id]
+
+    # A. flagged by IP only -> watch started with that baseline.
+    main.record_signup_ip("301", "watched1", "198.51.100.30", "datacenter", "Cloud Inc", True)
+    main.maybe_start_suspicious_watch("301", "watched1")
+    with main._db() as conn:
+        row = conn.execute(
+            "SELECT reasons, baseline_statuses_count, baseline_following_count, status "
+            "FROM suspicious_watch WHERE account_id = ?", ("301",),
+        ).fetchone()
+    assert row == ("ip", 2, 5, "pending"), row
+
+    # B. calling again (e.g. account.created then account.approved both
+    #    firing) does not reset/duplicate the baseline.
+    counts["301"] = (99, 99)
+    main.maybe_start_suspicious_watch("301", "watched1")
+    with main._db() as conn:
+        row = conn.execute(
+            "SELECT baseline_statuses_count FROM suspicious_watch WHERE account_id = ?", ("301",),
+        ).fetchone()
+    assert row == (2,), "re-calling must not overwrite the original baseline"
+
+    # C. never flagged -> no watch row at all.
+    main.maybe_start_suspicious_watch("302", "clean3")
+    with main._db() as conn:
+        row = conn.execute(
+            "SELECT 1 FROM suspicious_watch WHERE account_id = ?", ("302",),
+        ).fetchone()
+    assert row is None, "an unflagged signup must never be watched"
+
+
+def suspicious_sweep_tests():
+    """Drive app.suspicious_sweep.sweep() directly against rows seeded into
+    the shared sqlite store, with the Mastodon admin-account fetch and
+    apply_action/dm_moderator stubbed out (no network)."""
+    import app.suspicious_sweep as sweep_mod
+
+    sweep_mod.SUSPICIOUS_SWEEP_ENABLED = True
+    sweep_mod.SUSPICIOUS_GRACE_HOURS = 24
+    sweep_mod.SUSPICIOUS_ACTION = "suspend"
+    sweep_mod.SUSPICIOUS_DRY_RUN = False
+    sweep_mod.ABUSE_SKIP_PRIVILEGED = True
+    sweep_mod.ABUSE_ALLOWLIST = {"trustedstaff"}
+
+    def seed(account_id, acct, reasons, baseline_statuses, baseline_following, hours_ago):
+        started_at = (datetime.now(timezone.utc) - timedelta(hours=hours_ago)).isoformat()
+        with main._db() as conn:
+            conn.execute(
+                "INSERT INTO suspicious_watch "
+                "(account_id, acct, reasons, started_at, baseline_statuses_count, "
+                " baseline_following_count) VALUES (?, ?, ?, ?, ?, ?)",
+                (account_id, acct, reasons, started_at, baseline_statuses, baseline_following),
+            )
+
+    def status_of(account_id):
+        with main._db() as conn:
+            row = conn.execute(
+                "SELECT status FROM suspicious_watch WHERE account_id = ?", (account_id,),
+            ).fetchone()
+        return row[0] if row else None
+
+    admin_accounts = {}
+    sweep_mod._fetch_admin_account = lambda account_id: admin_accounts[account_id]
+
+    actions = []
+    dms = []
+    sweep_mod.apply_action = lambda target_id, action, text: actions.append((target_id, action))
+    sweep_mod.dm_moderator = lambda message: dms.append(message)
+
+    def public(statuses_count=0, following_count=0, **extra):
+        acct = {"statuses_count": statuses_count, "following_count": following_count}
+        return {"account": acct, **extra}
+
+    # A. not yet due (started 1h ago, 24h grace) -> untouched, still pending.
+    seed("401", "toosoon", "ip", 0, 0, hours_ago=1)
+    admin_accounts["401"] = public(0, 0)
+
+    # B. due, zero new posts/follows since baseline -> suspended.
+    seed("402", "dormantbot", "ip", 0, 0, hours_ago=25)
+    admin_accounts["402"] = public(0, 0)
+
+    # C. due, but posted since baseline -> cleared, not suspended.
+    seed("403", "postedsince", "email", 3, 1, hours_ago=25)
+    admin_accounts["403"] = public(4, 1)
+
+    # D. due, but followed someone since baseline -> cleared, not suspended.
+    seed("404", "followedsince", "ip,email", 0, 0, hours_ago=25)
+    admin_accounts["404"] = public(0, 1)
+
+    # E. due, but already suspended/silenced by another mechanism -> skipped.
+    seed("405", "alreadylimited", "ip", 0, 0, hours_ago=25)
+    admin_accounts["405"] = public(0, 0, suspended=True)
+
+    # F. due, but holds a staff role -> skipped, never suspended.
+    seed("406", "staffer", "ip", 0, 0, hours_ago=25)
+    admin_accounts["406"] = public(0, 0, role={"id": 3, "name": "Owner"})
+
+    # G. due, allowlisted by handle -> skipped.
+    seed("407", "trustedstaff", "ip", 0, 0, hours_ago=25)
+    admin_accounts["407"] = public(0, 0)
+
+    sweep_mod.sweep()
+
+    assert status_of("401") == "pending", "not-yet-due watch must be left alone"
+    assert ("402", "suspend") in actions, actions
+    assert status_of("402") == "suspend"
+    assert any("dormantbot" in d for d in dms), dms
+
+    assert ("403", "suspend") not in actions, actions
+    assert status_of("403") == "cleared"
+
+    assert ("404", "suspend") not in actions, actions
+    assert status_of("404") == "cleared"
+
+    assert ("405", "suspend") not in actions, actions
+    assert status_of("405") == "skipped-already-limited"
+
+    assert ("406", "suspend") not in actions, actions
+    assert status_of("406") == "skipped-privileged"
+
+    assert ("407", "suspend") not in actions, actions
+    assert status_of("407") == "skipped-privileged"
+
+    # H. re-running the sweep must not re-suspend or re-DM an already-resolved row.
+    before_actions, before_dms = len(actions), len(dms)
+    sweep_mod.sweep()
+    assert len(actions) == before_actions, "resolved rows must not be re-actioned"
+    assert len(dms) == before_dms, "resolved rows must not be re-DMed"
+
+    # I. dry-run mode logs/DMs but takes no real action.
+    seed("408", "dryrunbot", "ip", 0, 0, hours_ago=25)
+    admin_accounts["408"] = public(0, 0)
+    sweep_mod.SUSPICIOUS_DRY_RUN = True
+    sweep_mod.sweep()
+    assert ("408", "suspend") not in actions, "dry-run must not call apply_action"
+    assert status_of("408") == "dry-run:suspend"
+    assert any("[DRY-RUN]" in d and "dryrunbot" in d for d in dms), dms
+
+
 if __name__ == "__main__":
     routing_tests()
     policy_tests()
     ip_scrutiny_tests()
     email_scrutiny_tests()
+    suspicious_watch_tests()
+    suspicious_sweep_tests()
     print("ALL TESTS PASSED")