diff --git a/src/agent_reach/channels/rss.py b/src/agent_reach/channels/rss.py
index 2f08a6c..7f8f236 100644
--- a/src/agent_reach/channels/rss.py
+++ b/src/agent_reach/channels/rss.py
@@ -16,6 +16,7 @@ def fetch(command, query, params):
         raise ValueError("rss needs a feed URL")
     limit = int(params.get("limit", 20))
     parsed = feedparser.parse(query)
+    _check_status(parsed)
     if parsed.bozo and not parsed.entries:
         raise RuntimeError(f"could not read the feed: {parsed.bozo_exception}")
     items = []
@@ -37,6 +38,25 @@ def fetch(command, query, params):
     return items
 
 
+def _check_status(parsed):
+    # feedparser reports an HTTP 429 as bozo=False with no entries, which is
+    # indistinguishable from a feed that is genuinely empty. Raising here also
+    # keeps runner.run from caching the blocked response for the full TTL.
+    status = parsed.get("status")
+    if not isinstance(status, int) or status < 400:
+        return
+    if status == 429:
+        raise RuntimeError(
+            "rate limited (HTTP 429): the host is throttling this IP. Wait a few "
+            "minutes and retry. An empty result here means blocked, not quiet."
+        )
+    if status in (401, 403):
+        raise RuntimeError(f"the feed refused this request (HTTP {status})")
+    if status == 404:
+        raise RuntimeError(f"no feed at that URL (HTTP {status})")
+    raise RuntimeError(f"could not read the feed (HTTP {status})")
+
+
 def _strip_html(text):
     import html
     import re
diff --git a/tests/test_core.py b/tests/test_core.py
index 36e20b9..105bcc9 100644
--- a/tests/test_core.py
+++ b/tests/test_core.py
@@ -208,6 +208,47 @@ def test_rss_reads_a_local_feed():
     assert "escaped" in envelope.items[1].text
 
 
+def _fake_feed(**fields):
+    class FakeParsed(dict):
+        __getattr__ = dict.get
+
+    return FakeParsed({"bozo": False, "bozo_exception": None, "entries": [], **fields})
+
+
+def test_rss_raises_on_an_http_error_instead_of_returning_nothing(monkeypatch):
+    import feedparser
+
+    entry = manifest.get("rss")
+    for status, expected in [(429, "429"), (404, "404"), (403, "403"), (500, "500")]:
+        monkeypatch.setattr(feedparser, "parse", lambda url, s=status: _fake_feed(status=s))
+        with pytest.raises(RuntimeError, match=expected):
+            runner.run(entry, "feed", "https://example.com/feed.xml", use_cache=False)
+
+
+def test_rss_does_not_cache_a_blocked_response(monkeypatch):
+    import feedparser
+
+    entry = manifest.get("rss")
+    url = "https://example.com/feed.xml"
+    monkeypatch.setattr(feedparser, "parse", lambda _url: _fake_feed(status=429))
+    with pytest.raises(RuntimeError):
+        runner.run(entry, "feed", url)
+    assert cache.get(entry.name, "feed", url, {}, entry.cache_ttl) is None
+
+
+def test_rss_still_reads_a_feed_with_no_http_status(monkeypatch):
+    # feedparser sets no status key for a local path or raw bytes, so the guard
+    # must not mistake a missing status for a failure.
+    import feedparser
+
+    entry = manifest.get("rss")
+    monkeypatch.setattr(
+        feedparser, "parse", lambda _url: _fake_feed(entries=[{"title": "Local", "link": "", "summary": "hi"}])
+    )
+    envelope = runner.run(entry, "feed", "whatever", use_cache=False)
+    assert [i.title for i in envelope.items] == ["Local"]
+
+
 def test_vtt_parsing_drops_timestamps_and_repeats():
     raw = (FIXTURES / "sample.vtt").read_text()
     text = youtube.parse_vtt(raw)