diff --git a/conftest.py b/conftest.py
new file mode 100644
index 0000000..9a2f26e
--- /dev/null
+++ b/conftest.py
@@ -0,0 +1,128 @@
+"""Shared pytest fixtures for TinyWeb tests.
+
+Three fixtures cover most tests: `temp_db` swaps the SQLite path to a
+per-test tempfile, `seeded_db` layers sample rows on top, and `csrf_session`
+primes the thread-local CSRF token that handlers read.
+"""
+import socket
+import sys
+from pathlib import Path
+
+import pytest
+
+sys.path.insert(0, str(Path(__file__).parent))
+
+import db as db_module
+import handlers as handlers_module
+
+
+@pytest.fixture
+def temp_db(tmp_path, monkeypatch):
+    """Isolated SQLite DB per test.
+
+    Swaps `db.DATABASE` and `db.DATA_DIR` to a tempdir, clears the connection
+    pool before and after so state doesn't leak across tests, and calls
+    `init_db()` so every schema object exists.
+    """
+    data_dir = tmp_path / "tinyweb"
+    data_dir.mkdir()
+    db_path = data_dir / "index.db"
+
+    monkeypatch.setattr(db_module, "DATA_DIR", str(data_dir))
+    monkeypatch.setattr(db_module, "DATABASE", str(db_path))
+
+    with db_module._pool_lock:
+        for conn in db_module._pool:
+            try:
+                conn.close()
+            except Exception:
+                pass
+        db_module._pool.clear()
+
+    db_module.init_db()
+    yield db_path
+
+    with db_module._pool_lock:
+        for conn in db_module._pool:
+            try:
+                conn.close()
+            except Exception:
+                pass
+        db_module._pool.clear()
+
+
+@pytest.fixture
+def seeded_db(temp_db):
+    """A temp DB with a small, realistic set of pages/tags/links."""
+    db = db_module.get_db()
+    try:
+        rows = [
+            ("https://example.com/rust-intro", "Rust Intro", "A gentle introduction to rust borrow checker.", "notes on ownership"),
+            ("https://example.com/python-tips", "Python Tips", "Daily python tricks for readable code.", ""),
+            ("https://example.com/ocaml-why", "Why OCaml", "Type systems and inference in ocaml.", "private thoughts"),
+            ("https://news.example.org/mesh", "Mesh Networking", "Reticulum and LoRa for decentralized networks.", ""),
+        ]
+        for url, title, body, note in rows:
+            db.execute(
+                "INSERT INTO pages (url, title, body, note, last_modified) "
+                "VALUES (?, ?, ?, ?, '2026-04-01T00:00:00')",
+                (url, title, body, note),
+            )
+        db.commit()
+        page_ids = {
+            row["url"]: row["id"]
+            for row in db.execute("SELECT id, url FROM pages").fetchall()
+        }
+        tag_rows = [
+            (page_ids["https://example.com/rust-intro"], ["rust", "public"]),
+            (page_ids["https://example.com/python-tips"], ["python"]),
+            (page_ids["https://example.com/ocaml-why"], ["ocaml", "private"]),
+            (page_ids["https://news.example.org/mesh"], ["mesh", "public"]),
+        ]
+        for pid, tags in tag_rows:
+            for name in tags:
+                db.execute("INSERT OR IGNORE INTO tags (name) VALUES (?)", (name,))
+                tid = db.execute("SELECT id FROM tags WHERE name = ?", (name,)).fetchone()[0]
+                db.execute(
+                    "INSERT OR IGNORE INTO page_tags (page_id, tag_id) VALUES (?, ?)",
+                    (pid, tid),
+                )
+        db.execute(
+            "INSERT INTO links (page_id, url, label) VALUES (?, ?, ?)",
+            (page_ids["https://example.com/rust-intro"], "https://example.com/rust-advanced", "advanced rust guide"),
+        )
+        db.commit()
+    finally:
+        db_module.return_db(db)
+    return temp_db
+
+
+@pytest.fixture
+def csrf_session(monkeypatch):
+    """Prime the CSRF thread-local so handler code that calls _get_csrf_token works."""
+    token = "test-csrf-token"
+    handlers_module._request_local.csrf_token = token
+    yield token
+    if hasattr(handlers_module._request_local, "csrf_token"):
+        del handlers_module._request_local.csrf_token
+
+
+def patch_dns_fail(monkeypatch):
+    """Make every socket.getaddrinfo call raise gaierror for the rest of this test."""
+    def boom(*args, **kwargs):
+        raise socket.gaierror("test: DNS disabled")
+    monkeypatch.setattr(socket, "getaddrinfo", boom)
+
+
+def patch_dns_ok(monkeypatch, address="93.184.216.34"):
+    """Make every getaddrinfo return a single public IP for the rest of this test."""
+    def ok(host, port, *args, **kwargs):
+        return [(socket.AF_INET, socket.SOCK_STREAM, 0, "", (address, port or 80))]
+    monkeypatch.setattr(socket, "getaddrinfo", ok)
+
+
+def patch_dns_private(monkeypatch, address="127.0.0.1"):
+    """Make every getaddrinfo return a private/blocked IP for the rest of this test."""
+    def private(host, port, *args, **kwargs):
+        return [(socket.AF_INET, socket.SOCK_STREAM, 0, "", (address, port or 80))]
+    monkeypatch.setattr(socket, "getaddrinfo", private)
diff --git a/pytest.ini b/pytest.ini
new file mode 100644
index 0000000..16d6cc5
--- /dev/null
+++ b/pytest.ini
@@ -0,0 +1,5 @@
+[pytest]
+testpaths = tests
+python_files = test_*.py
+filterwarnings =
+    ignore::DeprecationWarning
diff --git a/requirements-dev.txt b/requirements-dev.txt
new file mode 100644
index 0000000..26b77f6
--- /dev/null
+++ b/requirements-dev.txt
@@ -0,0 +1,2 @@
+-r requirements.txt
+pytest
diff --git a/tests/test_csrf.py b/tests/test_csrf.py
new file mode 100644
index 0000000..43b4487
--- /dev/null
+++ b/tests/test_csrf.py
@@ -0,0 +1,60 @@
+"""Tests for `_check_csrf` — form-submission CSRF protection.
+
+Every POST handler calls this to verify the submitted _csrf field matches
+the token stored in the thread-local (which is seeded from the cookie by
+`dispatch_request`). Missing or mismatched tokens must fail closed.
+"""
+import handlers as handlers_module
+from handlers import _check_csrf, _csrf_field, _get_csrf_token
+
+
+def _set_token(token):
+    handlers_module._request_local.csrf_token = token
+
+
+def _clear_token():
+    if hasattr(handlers_module._request_local, "csrf_token"):
+        del handlers_module._request_local.csrf_token
+
+
+def teardown_function(_):
+    _clear_token()
+
+
+def test_rejects_missing_token_in_body():
+    _set_token("server-side-token")
+    assert _check_csrf({}) is False
+
+
+def test_rejects_empty_token_in_body():
+    _set_token("server-side-token")
+    assert _check_csrf({"_csrf": [""]}) is False
+
+
+def test_rejects_mismatched_token():
+    _set_token("server-side-token")
+    assert _check_csrf({"_csrf": ["attacker-token"]}) is False
+
+
+def test_accepts_matching_token():
+    _set_token("server-side-token")
+    assert _check_csrf({"_csrf": ["server-side-token"]}) is True
+
+
+def test_rejects_when_server_token_missing():
+    """If the server-side token is empty (shouldn't happen after dispatch_request
+    seeds it, but be defensive), the check must fail closed."""
+    _clear_token()
+    assert _check_csrf({"_csrf": ["anything"]}) is False
+
+
+def test_csrf_field_renders_current_token():
+    _set_token("abc123")
+    field = _csrf_field()
+    assert 'name="_csrf"' in field
+    assert 'value="abc123"' in field
+
+
+def test_get_csrf_token_returns_empty_when_unset():
+    _clear_token()
+    assert _get_csrf_token() == ""
diff --git a/tests/test_db_index_url.py b/tests/test_db_index_url.py
new file mode 100644
index 0000000..50f73ce
--- /dev/null
+++ b/tests/test_db_index_url.py
@@ -0,0 +1,155 @@
+"""Tests for `index_url` — the main write path.
+
+Covers UPSERT behavior, links being replaced on re-index, FTS index staying
+in sync via triggers, and the connection pool returning clean connections.
+"""
+from unittest.mock import patch
+
+from conftest import patch_dns_ok
+import db as db_module
+from db import get_db, return_db, index_url
+
+
+def _mock_fetch_page(title="Test Page", body="test body text", links=None, meta=""):
+    """Return a replacement for db.fetch_page that yields canned data."""
+    links = links or []
+    def fake(url):
+        return (title, body, links, meta)
+    return fake
+
+
+def test_insert_creates_page_row_and_fts_entry(temp_db, monkeypatch):
+    patch_dns_ok(monkeypatch)
+    monkeypatch.setattr(db_module, "fetch_page", _mock_fetch_page(
+        title="Rust Intro", body="ownership and borrowing basics", links=[],
+    ))
+    index_url("https://example.com/rust")
+
+    db = get_db()
+    try:
+        row = db.execute("SELECT id, title, body FROM pages").fetchone()
+        assert row is not None
+        assert row["title"] == "Rust Intro"
+        assert "ownership" in row["body"]
+        # Verify FTS trigger fired.
+        fts_hits = db.execute(
+            "SELECT rowid FROM pages_fts WHERE pages_fts MATCH 'ownership*'"
+        ).fetchall()
+        assert len(fts_hits) == 1
+        assert fts_hits[0]["rowid"] == row["id"]
+    finally:
+        return_db(db)
+
+
+def test_re_indexing_same_url_updates_in_place(temp_db, monkeypatch):
+    patch_dns_ok(monkeypatch)
+    monkeypatch.setattr(db_module, "fetch_page", _mock_fetch_page(
+        title="First Title", body="first body", links=[],
+    ))
+    index_url("https://example.com/page")
+
+    monkeypatch.setattr(db_module, "fetch_page", _mock_fetch_page(
+        title="Second Title", body="second body", links=[],
+    ))
+    index_url("https://example.com/page")
+
+    db = get_db()
+    try:
+        rows = db.execute("SELECT title, body FROM pages").fetchall()
+    finally:
+        return_db(db)
+    assert len(rows) == 1, "re-indexing should UPDATE not INSERT"
+    assert rows[0]["title"] == "Second Title"
+
+
+def test_links_replaced_on_reindex(temp_db, monkeypatch):
+    patch_dns_ok(monkeypatch)
+    monkeypatch.setattr(db_module, "fetch_page", _mock_fetch_page(
+        title="T", body="b",
+        links=[("https://example.com/a", "first"), ("https://example.com/b", "second")],
+    ))
+    index_url("https://example.com/src")
+
+    monkeypatch.setattr(db_module, "fetch_page", _mock_fetch_page(
+        title="T", body="b",
+        links=[("https://example.com/c", "third-only")],
+    ))
+    index_url("https://example.com/src")
+
+    db = get_db()
+    try:
+        rows = db.execute("SELECT url FROM links").fetchall()
+    finally:
+        return_db(db)
+    urls = {r["url"] for r in rows}
+    assert urls == {"https://example.com/c"}, "old links should be deleted on reindex"
+
+
+def test_url_cleaned_before_insert(temp_db, monkeypatch):
+    """index_url should apply clean_url before touching the DB, so tracking params
+    don't create duplicate rows."""
+    patch_dns_ok(monkeypatch)
+    monkeypatch.setattr(db_module, "fetch_page", _mock_fetch_page(title="T", body="b"))
+    index_url("https://example.com/page?utm_source=twitter#frag")
+
+    db = get_db()
+    try:
+        rows = db.execute("SELECT url FROM pages").fetchall()
+    finally:
+        return_db(db)
+    assert len(rows) == 1
+    assert rows[0]["url"] == "https://example.com/page"
+
+
+def test_summary_populated_from_meta_description(temp_db, monkeypatch):
+    patch_dns_ok(monkeypatch)
+    long_meta = "A thoughtful description that exceeds twenty chars"
+    monkeypatch.setattr(db_module, "fetch_page", _mock_fetch_page(
+        title="T", body="b", meta=long_meta,
+    ))
+    index_url("https://example.com/page")
+
+    db = get_db()
+    try:
+        row = db.execute("SELECT summary FROM pages").fetchone()
+    finally:
+        return_db(db)
+    assert row["summary"] == long_meta
+
+
+def test_short_meta_description_not_stored_as_summary(temp_db, monkeypatch):
+    patch_dns_ok(monkeypatch)
+    monkeypatch.setattr(db_module, "fetch_page", _mock_fetch_page(
+        title="T", body="b", meta="too short",
+    ))
+    index_url("https://example.com/page")
+
+    db = get_db()
+    try:
+        row = db.execute("SELECT summary FROM pages").fetchone()
+    finally:
+        return_db(db)
+    assert row["summary"] == ""
+
+
+def test_pool_returns_clean_connection(temp_db, monkeypatch):
+    """Regression for 1bc695f — `return_db` should roll back uncommitted work
+    so the next consumer doesn't see stale state."""
+    patch_dns_ok(monkeypatch)
+    monkeypatch.setattr(db_module, "fetch_page", _mock_fetch_page(title="T", body="b"))
+    index_url("https://example.com/one")
+
+    # Take a connection, make a dirty uncommitted change, return it.
+    db = get_db()
+    db.execute("INSERT INTO pages (url, title, body) VALUES (?, ?, ?)",
+               ("https://dirty.example.com/", "dirty", "dirty"))
+    # NOTE: no commit here — this is the dirty state we want rolled back.
+    return_db(db)
+
+    # A later consumer must not see the dirty row.
+    db2 = get_db()
+    try:
+        urls = {r["url"] for r in db2.execute("SELECT url FROM pages").fetchall()}
+    finally:
+        return_db(db2)
+    assert "https://dirty.example.com/" not in urls
diff --git a/tests/test_db_schema.py b/tests/test_db_schema.py
new file mode 100644
index 0000000..5a4f77c
--- /dev/null
+++ b/tests/test_db_schema.py
@@ -0,0 +1,90 @@
+"""Tests for `init_db` and the settings key-value store.
+
+`init_db` is called unconditionally on startup, so it must be idempotent
+and create every table/trigger the rest of the app expects.
+"""
+from db import get_db, return_db, init_db, get_setting, set_setting, get_site_name
+
+
+EXPECTED_TABLES = {
+    "pages", "links", "settings", "subscriptions",
+    "remote_pages", "tags", "page_tags", "chunks",
+    # FTS5 virtual tables:
+    "pages_fts", "remote_pages_fts",
+}
+
+
+def test_all_expected_tables_exist(temp_db):
+    db = get_db()
+    try:
+        rows = db.execute(
+            "SELECT name FROM sqlite_master WHERE type IN ('table') AND name NOT LIKE 'sqlite_%'"
+        ).fetchall()
+        names = {r["name"] for r in rows}
+    finally:
+        return_db(db)
+    missing = EXPECTED_TABLES - names
+    assert not missing, f"tables missing after init_db: {missing}"
+
+
+def test_fts_triggers_exist(temp_db):
+    db = get_db()
+    try:
+        rows = db.execute(
+            "SELECT name FROM sqlite_master WHERE type = 'trigger'"
+        ).fetchall()
+        names = {r["name"] for r in rows}
+    finally:
+        return_db(db)
+    # These triggers keep pages_fts in sync with pages on insert/update/delete.
+    for trigger in ("pages_ai", "pages_ad", "pages_au"):
+        assert trigger in names, f"missing trigger {trigger}"
+
+
+def test_init_db_is_idempotent(temp_db):
+    """Running init_db twice on the same DB must not error or duplicate anything."""
+    init_db()
+    init_db()  # second call should be a no-op
+    db = get_db()
+    try:
+        count = db.execute(
+            "SELECT count(*) FROM sqlite_master WHERE name = 'pages'"
+        ).fetchone()[0]
+    finally:
+        return_db(db)
+    assert count == 1
+
+
+def test_get_setting_returns_default_when_missing(temp_db):
+    assert get_setting("nonexistent", "fallback") == "fallback"
+    assert get_setting("nonexistent") == ""
+
+
+def test_set_setting_then_get(temp_db):
+    set_setting("site_name", "my-personal-index")
+    assert get_setting("site_name") == "my-personal-index"
+
+
+def test_set_setting_updates_existing(temp_db):
+    set_setting("key", "first")
+    set_setting("key", "second")
+    assert get_setting("key") == "second"
+
+
+def test_get_site_name_has_default(temp_db):
+    assert get_site_name() == "tinyweb"
+
+
+def test_get_site_name_reflects_override(temp_db):
+    set_setting("site_name", "custom-site")
+    assert get_site_name() == "custom-site"
+
+
+def test_foreign_keys_pragma_enabled(temp_db):
+    """Pool connections should have foreign_keys=ON so CASCADE deletes work."""
+    db = get_db()
+    try:
+        row = db.execute("PRAGMA foreign_keys").fetchone()
+    finally:
+        return_db(db)
+    assert row[0] == 1
diff --git a/tests/test_fts_sanitizer.py b/tests/test_fts_sanitizer.py
new file mode 100644
index 0000000..ad061da
--- /dev/null
+++ b/tests/test_fts_sanitizer.py
@@ -0,0 +1,113 @@
+"""Tests for `_sanitize_fts_query`.
+
+The sanitizer is the boundary between user input and FTS5 MATCH syntax.
+Commit 1bc695f tightened it after noticing that colons and operator words
+could escape the quoting. These tests keep that regression dead.
+"""
+import pytest
+
+from handlers import _sanitize_fts_query
+
+
+def test_empty_query_returns_no_match_token():
+    assert _sanitize_fts_query("") == '""'
+    assert _sanitize_fts_query("   ") == '""'
+
+
+def test_single_word_becomes_prefix_match():
+    assert _sanitize_fts_query("rust") == "rust*"
+
+
+def test_multi_word_quotes_all_but_last():
+    result = _sanitize_fts_query("rust borrow checker")
+    assert result == '"rust" "borrow" checker*'
+
+
+def test_stopwords_are_dropped():
+    # "the" and "a" should vanish; only "cat" remains (and gets prefix star).
+    assert _sanitize_fts_query("the a cat") == "cat*"
+
+
+def test_all_stopwords_returns_no_match_token():
+    assert _sanitize_fts_query("the and or") == '""'
+
+
+@pytest.mark.parametrize("bad_char", ["'", "(", ")", "+", "-", "^", "~", ":"])
+def test_fts5_operators_stripped_from_tokens(bad_char):
+    """FTS5 special chars inside user tokens must not survive — regression for 1bc695f.
+
+    The sanitizer legitimately adds `"` around tokens and a trailing `*` for prefix
+    matching; both are excluded from this check.
+    """
+    payload = f"foo{bad_char}bar"
+    out = _sanitize_fts_query(payload)
+    assert bad_char not in out, f"{bad_char!r} leaked into {out!r}"
+
+
+def test_asterisk_only_appears_as_trailing_prefix():
+    """Input `*` should not become an in-token asterisk; the sanitizer's trailing `*` is fine."""
+    out = _sanitize_fts_query("foo*bar")
+    assert out.count("*") <= 1
+    if "*" in out:
+        assert out.endswith("*")
+
+
+def test_quote_in_input_does_not_break_out_of_quoted_token():
+    """A `"` in user input must not close the sanitizer's protective quoting.
+
+    The sanitizer wraps each non-last token in double quotes; if a stray `"` from
+    the user slipped through, the resulting FTS5 expression would be interpreted
+    as broken syntax or, worse, a column filter.
+    """
+    out = _sanitize_fts_query('foo"bar baz"qux')
+    # Each pair of quotes in the output should be balanced and around a clean token.
+    assert out.count('"') % 2 == 0
+    # No embedded quotes inside a quoted region.
+    import re
+    for match in re.findall(r'"[^"]*"', out):
+        inner = match[1:-1]
+        assert '"' not in inner
+
+
+@pytest.mark.parametrize("op", ["AND", "OR", "NOT", "NEAR", "and", "or", "not", "near"])
+def test_fts5_operator_words_dropped(op):
+    """AND/OR/NOT/NEAR would be interpreted as operators on the unquoted last token."""
+    out = _sanitize_fts_query(f"foo {op} bar")
+    # the operator word itself should not appear
+    assert op.upper() not in out.upper().split('"'), f"operator {op!r} survived in {out!r}"
+
+
+def test_injection_payload_produces_valid_fts5():
+    """End-to-end: a realistic injection payload must produce syntactically valid FTS5.
+
+    We run the sanitized output through a throwaway FTS5 table; if the sanitizer
+    leaks operator characters the MATCH either raises or interprets malicious syntax.
+    """
+    import sqlite3
+    conn = sqlite3.connect(":memory:")
+    conn.execute("CREATE VIRTUAL TABLE t USING fts5(body)")
+    conn.execute("INSERT INTO t (body) VALUES ('hello world')")
+
+    for payload in [
+        'foo": OR bar NOT baz AND qux*()',
+        '" OR 1=1 --',
+        "title:secret AND public",
+        "(((",
+        "^^^~~~",
+    ]:
+        q = _sanitize_fts_query(payload)
+        # Must not raise — if operators leaked, FTS5 would error or mis-parse.
+        conn.execute("SELECT * FROM t WHERE t MATCH ?", (q,)).fetchall()
+    conn.close()
+
+
+def test_whitespace_only_tokens_dropped():
+    # tokens that become empty after stripping special chars should not produce bare quotes
+    out = _sanitize_fts_query('""" "" ""')
+    assert out == '""'
+
+
+def test_colon_stripped():
+    """Regression for 1bc695f — colon is an FTS5 column filter and must be stripped."""
+    out = _sanitize_fts_query("title:secret")
+    assert ":" not in out
diff --git a/tests/test_gateway_limits.py b/tests/test_gateway_limits.py
new file mode 100644
index 0000000..6033c3a
--- /dev/null
+++ b/tests/test_gateway_limits.py
@@ -0,0 +1,164 @@
+"""Tests for gateway-level guards: body-size cap and Reticulum surface whitelist.
+
+Regression targets from commit 1bc695f — a 16 MiB upload limit (DoS guard)
+and a strict GET-/api/sites-only whitelist for requests arriving over the
+Reticulum mesh (CSRF can't protect mesh callers, so gate by whitelist).
+"""
+import io
+
+import pytest
+
+import app as app_module
+from gateway import GatewayHandler, MAX_BODY_SIZE
+
+
+class FakeHeaders:
+    """Minimal replacement for http.server request headers."""
+    def __init__(self, items=None):
+        self._items = dict(items or {})
+
+    def get(self, key, default=None):
+        return self._items.get(key, default)
+
+
+class FakeGatewayHandler(GatewayHandler):
+    """Bypass the socket-bound __init__ and capture response calls in memory."""
+    def __init__(self, path="/", method="POST", headers=None, rfile=None):
+        self.path = path
+        self.command = method
+        self.headers = FakeHeaders(headers or {})
+        self.rfile = rfile or io.BytesIO()
+        self.wfile = io.BytesIO()
+        self._captured = {
+            "error": None, "status": None, "headers": [], "body_written": None,
+        }
+
+    def send_error(self, code, msg=""):
+        self._captured["error"] = (code, msg)
+
+    def send_response(self, code):
+        self._captured["status"] = code
+
+    def send_header(self, k, v):
+        self._captured["headers"].append((k, v))
+
+    def end_headers(self):
+        pass
+
+
+def test_post_over_size_cap_rejected_with_413():
+    """Regression for 1bc695f: request bodies over MAX_BODY_SIZE must be rejected
+    without being read into memory."""
+    oversize = MAX_BODY_SIZE + 1
+    handler = FakeGatewayHandler(
+        path="/add",
+        method="POST",
+        headers={"Content-Length": str(oversize)},
+    )
+    handler._forward("POST")
+    assert handler._captured["error"] is not None
+    code, _msg = handler._captured["error"]
+    assert code == 413
+
+
+def test_post_at_size_cap_accepted():
+    """A body exactly at MAX_BODY_SIZE should not be rejected by the size check."""
+    handler = FakeGatewayHandler(
+        path="/_does_not_matter",
+        method="POST",
+        headers={"Content-Length": str(MAX_BODY_SIZE)},
+        # rfile has no data; handler will try to read; local_dispatch isn't set.
+        # We only care that the 413 check passes, not that the request succeeds.
+        rfile=io.BytesIO(b""),
+    )
+    # Stub out local_dispatch so _forward doesn't try the network path.
+    from gateway import GatewayState
+    original = GatewayState.local_dispatch
+    GatewayState.local_dispatch = lambda data: {
+        "status": 404, "content_type": "text/plain", "body": "nope",
+    }
+    try:
+        handler._forward("POST")
+    finally:
+        GatewayState.local_dispatch = original
+    # Not a 413, because the body is exactly at the cap (cap is inclusive).
+    if handler._captured["error"]:
+        assert handler._captured["error"][0] != 413
+
+
+def test_negative_content_length_rejected():
+    handler = FakeGatewayHandler(
+        path="/add",
+        method="POST",
+        headers={"Content-Length": "-1"},
+    )
+    handler._forward("POST")
+    assert handler._captured["error"] is not None
+    code, _msg = handler._captured["error"]
+    assert code == 400
+
+
+def test_invalid_content_length_rejected():
+    handler = FakeGatewayHandler(
+        path="/add",
+        method="POST",
+        headers={"Content-Length": "abc"},
+    )
+    handler._forward("POST")
+    assert handler._captured["error"] is not None
+    code, _msg = handler._captured["error"]
+    assert code == 400
+
+
+# -------- Reticulum mesh surface whitelist --------
+
+
+def test_mesh_rejects_non_api_sites_get():
+    """Regression for 1bc695f: remote mesh callers can only GET /api/sites."""
+    resp = app_module.rns_request_handler(
+        path="/tinyweb",
+        data={"method": "GET", "path": "/pages", "query": {}, "body": {}, "gateway_host": ""},
+        request_id="x", link_id="y", remote_identity=None, requested_at=0,
+    )
+    assert resp["status"] == 403
+
+
+def test_mesh_rejects_post_to_api_sites():
+    resp = app_module.rns_request_handler(
+        path="/tinyweb",
+        data={"method": "POST", "path": "/api/sites", "query": {}, "body": {}, "gateway_host": ""},
+        request_id="x", link_id="y", remote_identity=None, requested_at=0,
+    )
+    assert resp["status"] == 403
+
+
+def test_mesh_rejects_sensitive_local_endpoints():
+    for path in ("/add", "/delete/1", "/style", "/import", "/export"):
+        resp = app_module.rns_request_handler(
+            path="/tinyweb",
+            data={"method": "GET", "path": path, "query": {}, "body": {}, "gateway_host": ""},
+            request_id="x", link_id="y", remote_identity=None, requested_at=0,
+        )
+        assert resp["status"] == 403, f"path {path!r} leaked through mesh whitelist"
+
+
+def test_mesh_allows_api_sites_get(temp_db, csrf_session):
+    """Sanity check: the one whitelisted combination is accepted."""
+    resp = app_module.rns_request_handler(
+        path="/tinyweb",
+        data={"method": "GET", "path": "/api/sites", "query": {}, "body": {}, "gateway_host": ""},
+        request_id="x", link_id="y", remote_identity=None, requested_at=0,
+    )
+    # Status depends on handler output; 200 is the happy path.
+    assert resp["status"] in (200, 403)  # 403 if sharing is disabled by default
+
+
+def test_mesh_handles_missing_data_payload():
+    """Regression-minded check: a None or malformed data object shouldn't crash."""
+    resp = app_module.rns_request_handler(
+        path="/tinyweb",
+        data=None,
+        request_id="x", link_id="y", remote_identity=None, requested_at=0,
+    )
+    # Default data has method=GET, path=/ which is not in the whitelist.
+    assert resp["status"] == 403
diff --git a/tests/test_handlers_pages.py b/tests/test_handlers_pages.py
new file mode 100644
index 0000000..ab4704c
--- /dev/null
+++ b/tests/test_handlers_pages.py
@@ -0,0 +1,174 @@
+"""Tests for `handle_bulk_action`, edit flow, and the bulk-delete confirm step.
+
+The bulk-delete confirmation flow is a data-loss guard added in commit
+8dffd8c — a stray POST without `confirmed=1` must render the confirmation
+page instead of actually deleting.
+"""
+from db import get_db, return_db
+from handlers import (
+    handle_bulk_action,
+    handle_edit_form,
+    handle_edit_submit,
+    handle_pages,
+)
+
+
+def _all_urls(seeded_db):
+    db = get_db()
+    try:
+        return {r["url"] for r in db.execute("SELECT url FROM pages").fetchall()}
+    finally:
+        return_db(db)
+
+
+def _page_id(seeded_db, url):
+    db = get_db()
+    try:
+        return db.execute("SELECT id FROM pages WHERE url = ?", (url,)).fetchone()["id"]
+    finally:
+        return_db(db)
+
+
+def test_bulk_delete_without_confirmed_renders_confirm_page(seeded_db, csrf_session):
+    """Regression for 8dffd8c: bulk delete must NOT delete until confirmed=1 is set."""
+    pid = _page_id(seeded_db, "https://example.com/rust-intro")
+    urls_before = _all_urls(seeded_db)
+
+    resp = handle_bulk_action({
+        "ids": [str(pid)],
+        "action": ["delete"],
+    })
+    assert resp["status"] == 200
+    assert "confirm delete" in resp["body"].lower()
+    assert "Rust Intro" in resp["body"]
+    # Must still show a hidden confirmed=1 field in the follow-up form.
+    assert 'name="confirmed" value="1"' in resp["body"]
+
+    # Crucially: nothing should have been deleted.
+    assert _all_urls(seeded_db) == urls_before
+
+
+def test_bulk_delete_with_confirmed_actually_deletes(seeded_db, csrf_session):
+    pid = _page_id(seeded_db, "https://example.com/rust-intro")
+
+    resp = handle_bulk_action({
+        "ids": [str(pid)],
+        "action": ["delete"],
+        "confirmed": ["1"],
+    })
+    # Confirmed delete redirects back to /pages.
+    assert resp["status"] in (302, 303)
+
+    urls = _all_urls(seeded_db)
+    assert "https://example.com/rust-intro" not in urls
+    # Other pages untouched.
+    assert "https://example.com/python-tips" in urls
+
+
+def test_bulk_delete_with_no_ids_redirects(seeded_db, csrf_session):
+    resp = handle_bulk_action({
+        "ids": [],
+        "action": ["delete"],
+        "confirmed": ["1"],
+    })
+    assert resp["status"] in (302, 303)
+    assert _all_urls(seeded_db) == {
+        "https://example.com/rust-intro",
+        "https://example.com/python-tips",
+        "https://example.com/ocaml-why",
+        "https://news.example.org/mesh",
+    }
+
+
+def test_bulk_delete_rejects_non_integer_ids(seeded_db, csrf_session):
+    resp = handle_bulk_action({
+        "ids": ["not-a-number"],
+        "action": ["delete"],
+        "confirmed": ["1"],
+    })
+    assert resp["status"] == 400
+
+
+def test_bulk_retag_add_mode_merges_tags(seeded_db, csrf_session):
+    pid = _page_id(seeded_db, "https://example.com/python-tips")
+
+    handle_bulk_action({
+        "ids": [str(pid)],
+        "action": ["retag"],
+        "bulk_tags": ["scripting, tutorials"],
+        "tag_mode": ["add"],
+    })
+    db = get_db()
+    try:
+        rows = db.execute(
+            "SELECT t.name FROM tags t JOIN page_tags pt ON pt.tag_id = t.id "
+            "WHERE pt.page_id = ? ORDER BY t.name",
+            (pid,),
+        ).fetchall()
+    finally:
+        return_db(db)
+    tags = [r["name"] for r in rows]
+    assert "python" in tags  # existing kept
+    assert "scripting" in tags  # new added
+    assert "tutorials" in tags
+
+
+def test_bulk_retag_replace_mode_overwrites_tags(seeded_db, csrf_session):
+    pid = _page_id(seeded_db, "https://example.com/python-tips")
+
+    handle_bulk_action({
+        "ids": [str(pid)],
+        "action": ["retag"],
+        "bulk_tags": ["one, two"],
+        "tag_mode": ["replace"],
+    })
+    db = get_db()
+    try:
+        rows = db.execute(
+            "SELECT t.name FROM tags t JOIN page_tags pt ON pt.tag_id = t.id "
+            "WHERE pt.page_id = ?",
+            (pid,),
+        ).fetchall()
+    finally:
+        return_db(db)
+    tags = {r["name"] for r in rows}
+    assert tags == {"one", "two"}
+    assert "python" not in tags
+
+
+def test_edit_form_renders_current_values(seeded_db, csrf_session):
+    pid = _page_id(seeded_db, "https://example.com/rust-intro")
+    resp = handle_edit_form(pid)
+    assert resp["status"] == 200
+    assert "Rust Intro" in resp["body"]
+    # Existing tags should appear in the tag field.
+    assert "rust" in resp["body"]
+
+
+def test_edit_form_404_for_unknown_page(temp_db, csrf_session):
+    resp = handle_edit_form(99999)
+    assert resp["status"] == 404
+
+
+def test_edit_submit_updates_title_and_note(seeded_db, csrf_session):
+    pid = _page_id(seeded_db, "https://example.com/rust-intro")
+    handle_edit_submit(pid, {
+        "title": ["New Rust Title"],
+        "note": ["new annotation"],
+        "tags": ["rust, updated"],
+    })
+    db = get_db()
+    try:
+        row = db.execute("SELECT title, note FROM pages WHERE id = ?", (pid,)).fetchone()
+    finally:
+        return_db(db)
+    assert row["title"] == "New Rust Title"
+    assert row["note"] == "new annotation"
+
+
+def test_handle_pages_lists_indexed_pages(seeded_db, csrf_session):
+    resp = handle_pages({})
+    assert resp["status"] == 200
+    # Every seeded page title appears on the list page.
+    for title in ("Rust Intro", "Python Tips", "Why OCaml", "Mesh Networking"):
+        assert title in resp["body"]
diff --git a/tests/test_handlers_search.py b/tests/test_handlers_search.py
new file mode 100644
index 0000000..f7d2f9e
--- /dev/null
+++ b/tests/test_handlers_search.py
@@ -0,0 +1,63 @@
+"""Tests for `handle_search` — the home page + primary user flow."""
+from handlers import handle_search
+
+
+def test_empty_index_empty_query_shows_welcome(temp_db, csrf_session):
+    resp = handle_search({})
+    assert resp["status"] == 200
+    body = resp["body"]
+    assert "Your index is empty" in body
+    # Links the welcome panel offers as equal-weight starting points.
+    assert "/add" in body
+    assert "/style" in body
+    assert "/subscriptions" in body
+
+
+def test_empty_index_with_query_shows_no_results(temp_db, csrf_session):
+    resp = handle_search({"q": ["rust"]})
+    assert resp["status"] == 200
+    assert "No results in your index" in resp["body"]
+
+
+def test_populated_index_with_matching_query_returns_results(seeded_db, csrf_session):
+    resp = handle_search({"q": ["rust"]})
+    assert resp["status"] == 200
+    assert "Rust Intro" in resp["body"]
+    # Page count shown in meta line.
+    assert "4 pages indexed" in resp["body"]
+
+
+def test_query_only_matches_relevant_pages(seeded_db, csrf_session):
+    resp = handle_search({"q": ["ocaml"]})
+    body = resp["body"]
+    assert "Why OCaml" in body
+    assert "Python Tips" not in body
+    assert "Rust Intro" not in body
+
+
+def test_pagination_query_param_respected(seeded_db, csrf_session):
+    """A high page number should still render without crashing."""
+    resp = handle_search({"q": ["example"], "p": ["99"]})
+    assert resp["status"] == 200
+
+
+def test_trusted_sites_fallback_surfaces_when_query_matches_link_label(seeded_db, csrf_session):
+    """Links extracted from indexed pages act as a fallback when direct results
+    are absent or thin; labels are substring-matched case-insensitively."""
+    resp = handle_search({"q": ["advanced"]})
+    body = resp["body"]
+    # The label "advanced rust guide" is on a link extracted from rust-intro.
+    assert "advanced rust guide" in body
+    assert "trusted sites" in body
+
+
+def test_page_count_in_meta_line(seeded_db, csrf_session):
+    resp = handle_search({})
+    assert "4 pages indexed" in resp["body"]
+
+
+def test_csp_and_security_headers_not_in_handler_but_via_dispatch(seeded_db, csrf_session):
+    """Handler itself returns no security headers; dispatch_request wraps them.
+    This test documents the boundary so future refactors don't break assumptions."""
+    resp = handle_search({})
+    assert "headers" not in resp or "Content-Security-Policy" not in resp.get("headers", {})
diff --git a/tests/test_handlers_subs.py b/tests/test_handlers_subs.py
new file mode 100644
index 0000000..93ee97d
--- /dev/null
+++ b/tests/test_handlers_subs.py
@@ -0,0 +1,112 @@
+"""Tests for subscription handlers.
+
+Subscription add validates the destination hash (32-char hex) locally
+before calling `fetch_remote_sites`; browse uses cached remote_pages when
+available and falls back to a live fetch otherwise.
+"""
+from unittest.mock import patch
+
+import handlers as handlers_module
+from db import get_db, return_db
+from handlers import handle_subscription_add, handle_subscription_browse
+
+
+VALID_HASH = "a" * 32
+
+
+def _subscription_count():
+    db = get_db()
+    try:
+        return db.execute("SELECT count(*) FROM subscriptions").fetchone()[0]
+    finally:
+        return_db(db)
+
+
+def test_rejects_empty_dest_hash(temp_db, csrf_session):
+    resp = handle_subscription_add({"dest_hash": [""]})
+    assert "32-character" in resp["body"]
+    assert _subscription_count() == 0
+
+
+def test_rejects_wrong_length(temp_db, csrf_session):
+    resp = handle_subscription_add({"dest_hash": ["abc123"]})
+    assert "32-character" in resp["body"]
+    assert _subscription_count() == 0
+
+
+def test_rejects_non_hex(temp_db, csrf_session):
+    resp = handle_subscription_add({"dest_hash": ["z" * 32]})
+    assert "hex" in resp["body"].lower()
+    assert _subscription_count() == 0
+
+
+def test_rejects_unreachable_peer(temp_db, csrf_session):
+    with patch.object(handlers_module, "fetch_remote_sites") as fetch:
+        fetch.side_effect = ConnectionError("unreachable")
+        resp = handle_subscription_add({"dest_hash": [VALID_HASH]})
+    assert "Could not reach" in resp["body"]
+    assert _subscription_count() == 0
+
+
+def test_rejects_peer_with_sharing_disabled(temp_db, csrf_session):
+    with patch.object(handlers_module, "fetch_remote_sites") as fetch:
+        fetch.side_effect = PermissionError("sharing disabled")
+        resp = handle_subscription_add({"dest_hash": [VALID_HASH]})
+    assert "sharing disabled" in resp["body"]
+    assert _subscription_count() == 0
+
+
+def test_successful_add_records_subscription(temp_db, csrf_session):
+    with patch.object(handlers_module, "fetch_remote_sites") as fetch:
+        fetch.return_value = {"name": "alice", "sites": []}
+        resp = handle_subscription_add({"dest_hash": [VALID_HASH]})
+    assert "Subscribed to alice" in resp["body"]
+    assert _subscription_count() == 1
+
+
+def test_dest_hash_strips_angle_brackets(temp_db, csrf_session):
+    """Users often paste hashes as `<aaa...>` from RNS log output; strip them."""
+    with patch.object(handlers_module, "fetch_remote_sites") as fetch:
+        fetch.return_value = {"name": "bob", "sites": []}
+        resp = handle_subscription_add({"dest_hash": [f"<{VALID_HASH}>"]})
+    assert _subscription_count() == 1
+
+
+def test_browse_unknown_subscription_is_404(temp_db, csrf_session):
+    resp = handle_subscription_browse(99999)
+    assert resp["status"] == 404
+
+
+def test_browse_marks_already_indexed_urls(seeded_db, csrf_session):
+    # Insert a subscription + some remote pages (one duplicate of local, one new).
+    db = get_db()
+    try:
+        db.execute(
+            "INSERT INTO subscriptions (dest_hash, name) VALUES (?, ?)",
+            (VALID_HASH, "alice"),
+        )
+        sub_id = db.execute("SELECT id FROM subscriptions").fetchone()["id"]
+        db.execute(
+            "INSERT INTO remote_pages (subscription_id, url, title, note, tags) "
+            "VALUES (?, ?, ?, ?, ?)",
+            (sub_id, "https://example.com/rust-intro", "Alice rust pick", "", ""),
+        )
+        db.execute(
+            "INSERT INTO remote_pages (subscription_id, url, title, note, tags) "
+            "VALUES (?, ?, ?, ?, ?)",
+            (sub_id, "https://new.example.com/shiny", "Shiny New Link", "note", "tag1"),
+        )
+        db.commit()
+    finally:
+        return_db(db)
+
+    resp = handle_subscription_browse(sub_id)
+    body = resp["body"]
+    assert resp["status"] == 200
+    assert "already indexed" in body
+    # The duplicate URL should appear in the "already indexed" section.
+    assert "Alice rust pick" in body
+    # The new URL should be in the selectable section.
+    assert "Shiny New Link" in body
+    # Count summary: "2 site(s) available, 1 new"
+    assert "1 new" in body
diff --git a/tests/test_handlers_tags.py b/tests/test_handlers_tags.py
new file mode 100644
index 0000000..7ec8f05
--- /dev/null
+++ b/tests/test_handlers_tags.py
@@ -0,0 +1,101 @@
+"""Tests for tag helpers and the tag browse handler.
+
+Tags are stored via a join table, so orphaned rows in `tags` can accumulate
+if `_cleanup_orphaned_tags` isn't called after deletion/retagging. Tag
+counts shown in the UI rely on this being right.
+"""
+from db import get_db, return_db
+from handlers import (
+    _cleanup_orphaned_tags,
+    _get_page_tags,
+    _set_page_tags,
+    handle_tag_browse,
+    handle_tags,
+)
+
+
+def _page_id(url):
+    db = get_db()
+    try:
+        row = db.execute("SELECT id FROM pages WHERE url = ?", (url,)).fetchone()
+        return row["id"] if row else None
+    finally:
+        return_db(db)
+
+
+def _tag_names():
+    db = get_db()
+    try:
+        return {r["name"] for r in db.execute("SELECT name FROM tags").fetchall()}
+    finally:
+        return_db(db)
+
+
+def test_get_page_tags_returns_sorted_names(seeded_db):
+    pid = _page_id("https://example.com/rust-intro")
+    tags = _get_page_tags(pid)
+    assert tags == sorted(tags)  # alphabetical
+    assert "rust" in tags
+    assert "public" in tags
+
+
+def test_set_page_tags_replaces_existing(seeded_db):
+    pid = _page_id("https://example.com/rust-intro")
+    db = get_db()
+    try:
+        _set_page_tags(pid, "brand, new, tags", db)
+        db.commit()
+    finally:
+        return_db(db)
+    current = _get_page_tags(pid)
+    assert current == ["brand", "new", "tags"]
+
+
+def test_set_page_tags_splits_on_comma_and_lowercases(seeded_db):
+    pid = _page_id("https://example.com/python-tips")
+    db = get_db()
+    try:
+        _set_page_tags(pid, "Foo, BAR, baz", db)
+        db.commit()
+    finally:
+        return_db(db)
+    assert set(_get_page_tags(pid)) == {"foo", "bar", "baz"}
+
+
+def test_cleanup_orphaned_tags_removes_unreferenced(seeded_db):
+    # Clear all tags on one page; previously-unique tags become orphans.
+    pid = _page_id("https://example.com/rust-intro")
+    db = get_db()
+    try:
+        _set_page_tags(pid, "", db)  # empty string = no tags
+        # `rust` was only on the rust-intro page; `public` is also on mesh.
+        _cleanup_orphaned_tags(db)
+        db.commit()
+    finally:
+        return_db(db)
+    names = _tag_names()
+    assert "rust" not in names  # pruned
+    assert "public" in names  # still on mesh
+
+
+def test_handle_tag_browse_filters_by_tag(seeded_db, csrf_session):
+    resp = handle_tag_browse("rust", {})
+    assert resp["status"] == 200
+    body = resp["body"]
+    assert "Rust Intro" in body
+    assert "Python Tips" not in body
+    assert "Why OCaml" not in body
+
+
+def test_handle_tag_browse_unknown_tag_is_graceful(seeded_db, csrf_session):
+    resp = handle_tag_browse("no-such-tag", {})
+    # Should render a valid page with zero results, not error.
+    assert resp["status"] == 200
+
+
+def test_handle_tags_lists_all_tags_with_counts(seeded_db, csrf_session):
+    resp = handle_tags()
+    assert resp["status"] == 200
+    body = resp["body"]
+    for tag in ("rust", "python", "ocaml", "mesh", "public", "private"):
+        assert tag in body
diff --git a/tests/test_link_extraction.py b/tests/test_link_extraction.py
new file mode 100644
index 0000000..2d8c741
--- /dev/null
+++ b/tests/test_link_extraction.py
@@ -0,0 +1,138 @@
+"""Tests for link extraction inside `fetch_page`.
+
+Link extraction powers the "trusted sites" fallback on empty searches and
+feeds the `links` table. Rules: same-domain only, skip binary extensions,
+skip Wikipedia special pages, resolve relatives via urljoin.
+"""
+from unittest.mock import patch
+
+from conftest import patch_dns_ok
+import db as db_module
+
+
+class FakeResponse:
+    def __init__(self, text, status_code=200):
+        self.text = text
+        self.status_code = status_code
+        self.is_redirect = False
+        self.headers = {}
+
+    def raise_for_status(self):
+        if self.status_code >= 400:
+            raise Exception(f"status {self.status_code}")
+
+
+def _fetch_with_html(monkeypatch, url, html):
+    """Invoke fetch_page against `url` with `html` as the mocked response body."""
+    patch_dns_ok(monkeypatch)
+    with patch.object(db_module, "requests") as mock_requests:
+        mock_requests.get.return_value = FakeResponse(html)
+        return db_module.fetch_page(url)
+
+
+def test_only_same_domain_links_kept(monkeypatch):
+    html = """
+    <html><body>
+      <a href="https://example.com/a">same</a>
+      <a href="https://other.com/b">cross</a>
+      <a href="https://sub.example.com/c">subdomain</a>
+    </body></html>
+    """
+    _, _, links, _ = _fetch_with_html(monkeypatch, "https://example.com/", html)
+    urls = [u for u, _label in links]
+    assert "https://example.com/a" in urls
+    assert "https://other.com/b" not in urls
+    assert "https://sub.example.com/c" not in urls
+
+
+def test_binary_extensions_skipped(monkeypatch):
+    html = """
+    <html><body>
+      <a href="/real-page">keep</a>
+      <a href="/image.png">skip</a>
+      <a href="/doc.pdf">skip</a>
+      <a href="/archive.zip">skip</a>
+      <a href="/song.mp3">skip</a>
+      <a href="/styles.css">skip</a>
+    </body></html>
+    """
+    _, _, links, _ = _fetch_with_html(monkeypatch, "https://example.com/", html)
+    urls = [u for u, _label in links]
+    assert "https://example.com/real-page" in urls
+    for ext in (".png", ".pdf", ".zip", ".mp3", ".css"):
+        assert not any(u.endswith(ext) for u in urls), f"{ext} leaked through"
+
+
+def test_wikipedia_special_pages_skipped(monkeypatch):
+    html = """
+    <html><body>
+      <a href="/wiki/Main_Page">keep</a>
+      <a href="/wiki/Special:Random">skip</a>
+      <a href="/wiki/Talk:Foo">skip</a>
+      <a href="/wiki/User:Jimbo">skip</a>
+      <a href="/wiki/Category:Bar">skip</a>
+    </body></html>
+    """
+    _, _, links, _ = _fetch_with_html(monkeypatch, "https://example.com/", html)
+    urls = [u for u, _label in links]
+    assert "https://example.com/wiki/Main_Page" in urls
+    for skip in ("Special:Random", "Talk:Foo", "User:Jimbo", "Category:Bar"):
+        assert not any(skip in u for u in urls), f"wiki {skip!r} leaked"
+
+
+def test_relative_urls_resolved(monkeypatch):
+    html = """<html><body><a href="/relative/path">r</a></body></html>"""
+    _, _, links, _ = _fetch_with_html(monkeypatch, "https://example.com/start", html)
+    urls = [u for u, _label in links]
+    assert "https://example.com/relative/path" in urls
+
+
+def test_fragment_stripped_from_extracted_links(monkeypatch):
+    html = """<html><body><a href="/page#section">r</a></body></html>"""
+    _, _, links, _ = _fetch_with_html(monkeypatch, "https://example.com/", html)
+    urls = [u for u, _label in links]
+    assert "https://example.com/page" in urls
+    assert not any("#" in u for u in urls)
+
+
+def test_duplicate_links_deduped(monkeypatch):
+    html = """
+    <html><body>
+      <a href="/a">first</a>
+      <a href="/a">second</a>
+      <a href="/a">third</a>
+    </body></html>
+    """
+    _, _, links, _ = _fetch_with_html(monkeypatch, "https://example.com/", html)
+    urls = [u for u, _label in links]
+    assert urls.count("https://example.com/a") == 1
+
+
+def test_label_truncated_to_200(monkeypatch):
+    long_text = "x" * 500
+    html = f'<html><body><a href="/p">{long_text}</a></body></html>'
+    _, _, links, _ = _fetch_with_html(monkeypatch, "https://example.com/", html)
+    assert len(links) == 1
+    _, label = links[0]
+    assert len(label) <= 200
+
+
+def test_meta_description_extracted(monkeypatch):
+    html = """
+    <html><head>
+      <meta name="description" content="the real description">
+    </head><body><p>body content</p></body></html>
+    """
+    title, body, links, meta = _fetch_with_html(monkeypatch, "https://example.com/", html)
+    assert meta == "the real description"
+
+
+def test_og_description_fallback(monkeypatch):
+    """When there's no <meta name=description>, og:description wins."""
+    html = """
+    <html><head>
+      <meta property="og:description" content="open graph fallback">
+    </head><body><p>body</p></body></html>
+    """
+    _, _, _, meta = _fetch_with_html(monkeypatch, "https://example.com/", html)
+    assert meta == "open graph fallback"
diff --git a/tests/test_pagination.py b/tests/test_pagination.py
new file mode 100644
index 0000000..05077e0
--- /dev/null
+++ b/tests/test_pagination.py
@@ -0,0 +1,58 @@
+"""Tests for `_paginate` and `_page_nav`."""
+from handlers import _paginate, _page_nav, PER_PAGE
+
+
+def test_paginate_default_is_one():
+    assert _paginate({}) == 1
+
+
+def test_paginate_reads_query_string():
+    assert _paginate({"p": ["3"]}) == 3
+
+
+def test_paginate_clamps_to_one():
+    assert _paginate({"p": ["0"]}) == 1
+    assert _paginate({"p": ["-5"]}) == 1
+
+
+def test_paginate_handles_bad_input():
+    assert _paginate({"p": ["not-a-number"]}) == 1
+    assert _paginate({"p": []}) == 1
+
+
+def test_paginate_custom_key():
+    assert _paginate({"batch": ["7"]}, key="batch") == 7
+
+
+def test_page_nav_empty_when_single_page():
+    assert _page_nav(1, PER_PAGE, "/?q=foo") == ""
+    assert _page_nav(1, 0, "/?q=foo") == ""
+
+
+def test_page_nav_shows_next_on_first_page():
+    out = _page_nav(1, PER_PAGE * 3, "/?q=foo")
+    assert "next" in out
+    assert "prev" not in out
+    assert "page 1 of 3" in out
+
+
+def test_page_nav_shows_both_in_middle():
+    out = _page_nav(2, PER_PAGE * 3, "/?q=foo")
+    assert "next" in out
+    assert "prev" in out
+
+
+def test_page_nav_shows_prev_on_last_page():
+    out = _page_nav(3, PER_PAGE * 3, "/?q=foo")
+    assert "next" not in out
+    assert "prev" in out
+    assert "page 3 of 3" in out
+
+
+def test_page_nav_handles_query_string_separator():
+    # when base_url already has ?, pagination links must use &
+    out = _page_nav(1, PER_PAGE * 2, "/?q=foo")
+    assert "&p=2" in out
+    # when base_url has no ?, pagination links use ?
+    out = _page_nav(1, PER_PAGE * 2, "/pages")
+    assert "?p=2" in out
diff --git a/tests/test_regressions.py b/tests/test_regressions.py
new file mode 100644
index 0000000..f8a5df7
--- /dev/null
+++ b/tests/test_regressions.py
@@ -0,0 +1,107 @@
+"""Aggregator of regression tests tied to specific bug-fix commits.
+
+Each test here guards against a specific bug that was once shipped. Running
+just this file gives a one-line-per-bug audit:
+
+    pytest tests/test_regressions.py -v
+
+The test bodies are intentionally small; for the exhaustive behavior of each
+module, see the topical test files (test_fts_sanitizer.py, test_url_cleanup.py,
+etc.). This file's job is to make the bug catalog scannable.
+"""
+import socket
+from unittest.mock import patch
+
+import pytest
+
+import app as app_module
+import db as db_module
+import handlers as handlers_module
+from conftest import patch_dns_fail, patch_dns_ok
+from db import clean_url
+from handlers import _sanitize_fts_query, handle_bulk_action
+
+
+def test_6ffd38d_clean_url_preserves_www_when_bare_domain_fails(monkeypatch):
+    """6ffd38d: `clean_url` used to strip `www.` unconditionally; for sites that
+    only serve at `www.`, this produced unreachable clean URLs."""
+    patch_dns_fail(monkeypatch)
+    assert clean_url("https://www.example.com/page") == "https://www.example.com/page"
+
+
+def test_1bc695f_fts_sanitizer_strips_colon():
+    """1bc695f: FTS5 colon is a column filter — must not appear in sanitized output."""
+    assert ":" not in _sanitize_fts_query("title:secret body:exposed")
+
+
+@pytest.mark.parametrize("op", ["AND", "OR", "NOT", "NEAR"])
+def test_1bc695f_fts_sanitizer_drops_operator_words(op):
+    """1bc695f: operator words (AND/OR/NOT/NEAR) would be interpreted as FTS5
+    operators if they landed on the unquoted last token."""
+    out = _sanitize_fts_query(f"foo {op} bar")
+    # operator itself should not appear in the output
+    tokens = out.replace('"', '').split()
+    assert op not in [t.rstrip("*") for t in tokens]
+
+
+def test_1bc695f_gateway_rejects_oversize_body():
+    """1bc695f: 16 MiB body-size cap prevents memory-exhaustion DoS."""
+    from tests.test_gateway_limits import FakeGatewayHandler
+    from gateway import MAX_BODY_SIZE
+    h = FakeGatewayHandler(
+        path="/add", method="POST",
+        headers={"Content-Length": str(MAX_BODY_SIZE + 1)},
+    )
+    h._forward("POST")
+    assert h._captured["error"] and h._captured["error"][0] == 413
+
+
+def test_1bc695f_mesh_rejects_non_whitelisted_paths():
+    """1bc695f: Reticulum callers are limited to GET /api/sites; CSRF cannot
+    authenticate mesh callers."""
+    resp = app_module.rns_request_handler(
+        path="/tinyweb",
+        data={"method": "POST", "path": "/add", "query": {}, "body": {}, "gateway_host": ""},
+        request_id="x", link_id="y", remote_identity=None, requested_at=0,
+    )
+    assert resp["status"] == 403
+
+
+def test_1bc695f_pool_returns_clean_connection(temp_db, monkeypatch):
+    """1bc695f: uncommitted transactions on a pooled connection used to leak
+    into the next consumer."""
+    from db import get_db, return_db
+    db = get_db()
+    db.execute(
+        "INSERT INTO pages (url, title, body) VALUES (?, ?, ?)",
+        ("https://leak.example.com/", "should not persist", "body"),
+    )
+    return_db(db)  # no commit
+    db2 = get_db()
+    try:
+        urls = {r["url"] for r in db2.execute("SELECT url FROM pages").fetchall()}
+    finally:
+        return_db(db2)
+    assert "https://leak.example.com/" not in urls
+
+
+def test_8dffd8c_bulk_delete_requires_confirmation(seeded_db, csrf_session):
+    """8dffd8c: bulk delete without confirmed=1 must render a confirm page
+    instead of deleting — the JS confirm on /pages is a first-line filter only."""
+    from db import get_db, return_db
+    db = get_db()
+    try:
+        pid = db.execute("SELECT id FROM pages LIMIT 1").fetchone()["id"]
+        count_before = db.execute("SELECT count(*) FROM pages").fetchone()[0]
+    finally:
+        return_db(db)
+
+    resp = handle_bulk_action({"ids": [str(pid)], "action": ["delete"]})
+    assert "confirm delete" in resp["body"].lower()
+
+    db = get_db()
+    try:
+        count_after = db.execute("SELECT count(*) FROM pages").fetchone()[0]
+    finally:
+        return_db(db)
+    assert count_before == count_after, "bulk delete ran without confirmation"
diff --git a/tests/test_sharing_logic.py b/tests/test_sharing_logic.py
new file mode 100644
index 0000000..c9c06d4
--- /dev/null
+++ b/tests/test_sharing_logic.py
@@ -0,0 +1,38 @@
+"""Tests for `_page_is_shared`.
+
+This function decides whether a page is exposed over Reticulum to
+subscribers. Getting it wrong means either a privacy leak or silently
+hiding pages the user meant to share — both are worth a regression net.
+"""
+import pytest
+
+from handlers import _page_is_shared
+
+
+@pytest.mark.parametrize("mode", ["exclude_private", "require_public"])
+def test_private_tag_always_excludes(mode):
+    """`private` tag overrides every mode — the most important invariant."""
+    assert _page_is_shared(["private"], mode) is False
+    assert _page_is_shared(["public", "private"], mode) is False
+
+
+def test_exclude_private_defaults_to_shared():
+    assert _page_is_shared([], "exclude_private") is True
+    assert _page_is_shared(["random-tag"], "exclude_private") is True
+
+
+def test_require_public_needs_public_tag():
+    assert _page_is_shared([], "require_public") is False
+    assert _page_is_shared(["rust"], "require_public") is False
+    assert _page_is_shared(["public"], "require_public") is True
+
+
+def test_require_public_still_vetoes_private():
+    # public AND private → private wins.
+    assert _page_is_shared(["public", "private"], "require_public") is False
+
+
+def test_unknown_mode_treated_as_exclude_private():
+    """The default mode is 'exclude_private'; unknown modes fall through to it."""
+    assert _page_is_shared([], "totally-bogus-mode") is True
+    assert _page_is_shared(["private"], "totally-bogus-mode") is False
diff --git a/tests/test_ssrf.py b/tests/test_ssrf.py
new file mode 100644
index 0000000..807f9bd
--- /dev/null
+++ b/tests/test_ssrf.py
@@ -0,0 +1,64 @@
+"""Tests for `_validate_url_target` — SSRF prevention.
+
+Any URL the app fetches must resolve to a public IP; private/internal/
+loopback addresses must be rejected so attacker-controlled URLs cannot
+reach internal services via our HTTP client.
+"""
+import socket
+from unittest.mock import patch
+
+import pytest
+
+from db import _validate_url_target
+
+
+def _mock_getaddrinfo(address):
+    """Return a function suitable as a socket.getaddrinfo replacement."""
+    def f(host, port, *args, **kwargs):
+        family = socket.AF_INET6 if ":" in address else socket.AF_INET
+        return [(family, socket.SOCK_STREAM, 0, "", (address, port or 80))]
+    return f
+
+
+@pytest.mark.parametrize("blocked_ip", [
+    "127.0.0.1",
+    "127.1.2.3",
+    "10.0.0.1",
+    "10.255.255.255",
+    "172.16.0.1",
+    "172.31.255.255",
+    "192.168.0.1",
+    "192.168.255.255",
+    "169.254.169.254",
+    "0.0.0.0",
+    "::1",
+    "fc00::1",
+    "fe80::1",
+])
+def test_blocks_private_and_loopback(monkeypatch, blocked_ip):
+    monkeypatch.setattr(socket, "getaddrinfo", _mock_getaddrinfo(blocked_ip))
+    with pytest.raises(ValueError, match="blocked"):
+        _validate_url_target("https://evil.example.com/internal")
+
+
+def test_allows_public_ipv4(monkeypatch):
+    monkeypatch.setattr(socket, "getaddrinfo", _mock_getaddrinfo("8.8.8.8"))
+    _validate_url_target("https://dns.example.com/")  # does not raise
+
+
+def test_allows_public_ipv6(monkeypatch):
+    monkeypatch.setattr(socket, "getaddrinfo", _mock_getaddrinfo("2001:4860:4860::8888"))
+    _validate_url_target("https://v6.example.com/")  # does not raise
+
+
+def test_rejects_unresolvable_hostname(monkeypatch):
+    def boom(*args, **kwargs):
+        raise socket.gaierror("no such host")
+    monkeypatch.setattr(socket, "getaddrinfo", boom)
+    with pytest.raises(ValueError, match="Cannot resolve"):
+        _validate_url_target("https://does-not-exist.example.com/")
+
+
+def test_rejects_missing_hostname():
+    with pytest.raises(ValueError, match="No hostname"):
+        _validate_url_target("http:///path-only")
diff --git a/tests/test_url_cleanup.py b/tests/test_url_cleanup.py
new file mode 100644
index 0000000..1eef72b
--- /dev/null
+++ b/tests/test_url_cleanup.py
@@ -0,0 +1,101 @@
+"""Tests for `clean_url` — URL normalization and tracking-param stripping.
+
+Clean URLs are the deduplication key in the pages table, so any change to
+this function can silently cause duplicate rows or mask legitimate saves.
+"""
+import pytest
+
+from conftest import patch_dns_ok, patch_dns_fail
+from db import clean_url, TRACKING_PARAMS
+
+
+def test_strips_fragment(monkeypatch):
+    patch_dns_ok(monkeypatch)
+    assert clean_url("https://example.com/page#section") == "https://example.com/page"
+
+
+def test_prefers_https(monkeypatch):
+    patch_dns_ok(monkeypatch)
+    assert clean_url("http://example.com/page") == "https://example.com/page"
+
+
+def test_lowercases_hostname(monkeypatch):
+    patch_dns_ok(monkeypatch)
+    assert clean_url("https://EXAMPLE.COM/page") == "https://example.com/page"
+
+
+def test_preserves_path_case(monkeypatch):
+    """Paths are case-sensitive and should not be lowercased."""
+    patch_dns_ok(monkeypatch)
+    assert clean_url("https://example.com/Foo/Bar") == "https://example.com/Foo/Bar"
+
+
+def test_strips_default_https_port(monkeypatch):
+    patch_dns_ok(monkeypatch)
+    assert clean_url("https://example.com:443/page") == "https://example.com/page"
+
+
+@pytest.mark.xfail(reason="clean_url upgrades http->https before the port-default check, "
+                          "so port 80 is not stripped. Minor dedup bug — harmless but worth fixing.")
+def test_strips_http_port_80(monkeypatch):
+    """Expected: http://foo:80 → https://foo (both scheme-upgrade and port-strip).
+
+    Currently fails because scheme is upgraded to https *before* the port check,
+    so `scheme == "http" and port == 80` is never true by the time the check runs.
+    """
+    patch_dns_ok(monkeypatch)
+    assert clean_url("http://example.com:80/page") == "https://example.com/page"
+
+
+def test_preserves_non_default_port(monkeypatch):
+    patch_dns_ok(monkeypatch)
+    assert clean_url("https://example.com:8443/page") == "https://example.com:8443/page"
+
+
+def test_strips_trailing_slash(monkeypatch):
+    patch_dns_ok(monkeypatch)
+    assert clean_url("https://example.com/page/") == "https://example.com/page"
+
+
+def test_root_slash_preserved(monkeypatch):
+    patch_dns_ok(monkeypatch)
+    assert clean_url("https://example.com/") == "https://example.com/"
+
+
+@pytest.mark.parametrize("param", sorted(TRACKING_PARAMS))
+def test_tracking_params_stripped(monkeypatch, param):
+    patch_dns_ok(monkeypatch)
+    result = clean_url(f"https://example.com/page?{param}=value&keep=yes")
+    assert param not in result
+    assert "keep=yes" in result
+
+
+def test_strips_www_when_nonwww_resolves(monkeypatch):
+    """Standard case: strip `www.` prefix to canonicalize."""
+    patch_dns_ok(monkeypatch)
+    assert clean_url("https://www.example.com/page") == "https://example.com/page"
+
+
+def test_preserves_www_when_nonwww_does_not_resolve(monkeypatch):
+    """Regression for 6ffd38d.
+
+    Some sites only serve their content at `www.domain.tld`; the bare domain
+    doesn't resolve. Stripping `www.` in that case produced a URL that we could
+    never actually fetch or dedupe against the real one.
+    """
+    patch_dns_fail(monkeypatch)
+    assert clean_url("https://www.example.com/page") == "https://www.example.com/page"
+
+
+def test_query_params_sorted_for_stable_ordering(monkeypatch):
+    """Same URL with different param orderings should produce the same clean URL."""
+    patch_dns_ok(monkeypatch)
+    a = clean_url("https://example.com/page?b=2&a=1")
+    b = clean_url("https://example.com/page?a=1&b=2")
+    assert a == b
+
+
+def test_path_and_query_preserved_through_cleanup(monkeypatch):
+    patch_dns_ok(monkeypatch)
+    result = clean_url("https://example.com/path/to/page?id=42&utm_source=twitter")
+    assert result == "https://example.com/path/to/page?id=42"