diff --git a/crates/core/src/feed.rs b/crates/core/src/feed.rs
index 5e66acb5..0a790c5c 100644
--- a/crates/core/src/feed.rs
+++ b/crates/core/src/feed.rs
@@ -90,7 +90,7 @@ pub struct Notice {
     pub id: String,
     pub title: String,
     pub body: String,
-    /// Only `https://` / `http://` links survive parsing.
+    /// Only `https://` links survive parsing.
     pub link: Option<String>,
     pub created_at: Option<i64>,
 }
@@ -120,7 +120,8 @@ impl Feed {
             alerts: raw.important_notices.alerts.into_iter().filter_map(|n| n.into_notice("alert")).take(MAX_NOTICES).collect(),
             mirror: None,
         };
-        // App renames are news too: show them with the announcements.
+        // App renames are news too: show them with the announcements, inside the cap.
+        feed.announcements.truncate(MAX_NOTICES.saturating_sub(feed.app_renames.len()));
         for r in &feed.app_renames {
             feed.announcements.push(Notice {
                 id: format!("rename:{}:{}", r.old_name_lower, r.new_name_lower),
@@ -130,6 +131,7 @@ impl Feed {
                 created_at: r.renamed_at,
             });
         }
+        feed.announcements.truncate(MAX_NOTICES);
         Some(feed)
     }
 
@@ -183,9 +185,9 @@ impl Feed {
 
 impl FeedRelease {
     /// The release with only the downloads that belong to `repo` (or sit under `mirror`). `None`
-    /// when it doesn't hold up (no tag, or not a version).
+    /// when it doesn't hold up (no tag, a tag with odd characters, or not a version).
     fn to_release(&self, repo: &str, mirror: Option<&str>) -> Option<Release> {
-        let version = version::parse(&self.tag)?;
+        let version = version::parse(&self.tag).filter(|_| release::valid_tag(&self.tag))?;
         let allowed = |name: &str, url: &str| {
             (release::download_url_under(url, &release::download_prefix(repo))
                 && url == format!("{}{}/{}", release::download_prefix(repo), self.tag, name))
@@ -415,13 +417,7 @@ impl RawNotice {
         if self.title.trim().is_empty() && self.body.trim().is_empty() {
             return None;
         }
-        // The link may reach the system opener (whose WSL fallback shells out), so refuse any
-        // shell/cmd metacharacter or control character in it as well as whitespace (CWE-78).
-        let link = self.maybe_web_link.filter(|l| {
-            (l.starts_with("https://") || l.starts_with("http://"))
-                && l.len() <= 2000
-                && !l.chars().any(|c| c.is_whitespace() || c.is_control() || release::is_shell_meta(c))
-        });
+        let link = self.maybe_web_link.filter(|l| release::safe_web_link(l));
         Some(Notice {
             id: format!("{kind}:{}:{}", self.created_at.as_deref().unwrap_or(""), self.title.trim()),
             title: truncate(self.title.trim(), 200),
@@ -633,7 +629,10 @@ mod tests {
                 r#""alerts": [
             {"title": "Security update", "body": "Update PdfCraft now.", "maybe_web_link": "https://getartcraft.com/security", "created_at": null},
             {"title": "Bad link", "body": "x", "maybe_web_link": "javascript:alert(1)"},
-            {"title": "Shell link", "body": "x", "maybe_web_link": "https://getartcraft.com/a|calc.exe"}]"#,
+            {"title": "Shell link", "body": "x", "maybe_web_link": "https://getartcraft.com/a|calc.exe"},
+            {"title": "Plain http", "body": "x", "maybe_web_link": "http://getartcraft.com/security"},
+            {"title": "User-info", "body": "x", "maybe_web_link": "https://getartcraft.com@evil.example/"},
+            {"title": "File link", "body": "x", "maybe_web_link": "file:///etc/passwd"}]"#,
             )
             .replace(
                 "\"app_renames\": []",
@@ -641,13 +640,44 @@ mod tests {
             "new_name_lower": "docucraft", "description": "Same app, new name.", "renamed_at": "2026-10-09T08:00:00Z"}]"#,
             );
         let feed = Feed::parse(&with_notices).unwrap();
-        assert_eq!(feed.alerts.len(), 3);
+        assert_eq!(feed.alerts.len(), 6);
         assert_eq!(feed.alerts[0].link.as_deref(), Some("https://getartcraft.com/security"));
         assert_eq!(feed.alerts[1].link, None, "only web links are kept");
         assert_eq!(feed.alerts[2].link, None, "a link with shell metacharacters is dropped");
+        assert_eq!(feed.alerts[3].link, None, "plain http is dropped");
+        assert_eq!(feed.alerts[4].link, None, "a link with user-info is dropped");
+        assert_eq!(feed.alerts[5].link, None, "a file link is dropped");
         let titles: Vec<&str> = feed.announcements.iter().map(|n| n.title.as_str()).collect();
         assert_eq!(titles, ["New: CADCraft 1.0", "PdfCraft is now DocuCraft"]);
         // Ids are stable across parses, so dismissals stick.
         assert_eq!(Feed::parse(&with_notices).unwrap().alerts[0].id, feed.alerts[0].id);
     }
+
+    #[test]
+    fn rename_notices_count_toward_the_cap() {
+        let rename = |i: usize| {
+            format!(
+                r#"{{"old_name": "PdfCraft{i}", "old_name_lower": "pdfcraft{i}", "new_name": "DocuCraft{i}", "new_name_lower": "docucraft{i}",
+            "description": "Same app, new name.", "renamed_at": "2026-10-09T08:00:00Z"}}"#
+            )
+        };
+        for (news, renames) in [(MAX_NOTICES, 1), (1000, 1), (0, 30)] {
+            let many: Vec<String> = (0..news).map(|i| format!(r#"{{"title": "News {i}", "body": "x"}}"#)).collect();
+            let moved: Vec<String> = (0..renames).map(rename).collect();
+            let full = sample()
+                .replace("\"announcements\": []", &format!("\"announcements\": [{}]", many.join(",")))
+                .replace("\"app_renames\": []", &format!("\"app_renames\": [{}]", moved.join(",")));
+            let notices = Feed::parse(&full).unwrap().announcements;
+            assert_eq!(notices.len(), MAX_NOTICES, "{news} announcements, {renames} renames");
+            assert!(notices.iter().any(|n| n.title == "PdfCraft0 is now DocuCraft0"), "{news} announcements: the rename was cut");
+        }
+    }
+
+    #[test]
+    fn a_feed_release_with_an_odd_tag_asks_github() {
+        for odd in ["v0.6.0~beta-\\u202e", "v0.6.0~beta-a/b"] {
+            let feed = sample().replace("\"tag\": \"v0.6.0\"", &format!("\"tag\": \"{odd}\""));
+            assert_eq!(Feed::parse(&feed).unwrap().answer(app("photocraft"), NOW), FeedAnswer::AskGitHub, "{odd:?}");
+        }
+    }
 }
diff --git a/crates/core/src/release.rs b/crates/core/src/release.rs
index f599b999..f9b30557 100644
--- a/crates/core/src/release.rs
+++ b/crates/core/src/release.rs
@@ -143,7 +143,7 @@ fn download_suffix_ok(rest: &str) -> bool {
     }
     let mut parts = rest.split('/');
     let (Some(tag), Some(name), None) = (parts.next(), parts.next(), parts.next()) else { return false };
-    !tag.is_empty() && tag != "." && tag != ".." && tag.len() <= 64 && !tag.contains('\\') && valid_file_name(name)
+    valid_tag(tag) && valid_file_name(name)
 }
 
 /// Any `.` or `..` path segment that could climb out of the repository.
@@ -228,7 +228,7 @@ fn release_from_value(v: &serde_json::Value, repo: &str) -> Result<Release, Stri
     if let Some(msg) = v["message"].as_str() {
         return Err(if msg == "Not Found" { "no releases yet".into() } else { format!("GitHub says: {}", truncate(msg, 200)) });
     }
-    let tag = v["tag_name"].as_str().filter(|t| !t.is_empty() && t.len() <= 64).ok_or("no release found")?.to_string();
+    let tag = v["tag_name"].as_str().filter(|t| valid_tag(t)).ok_or("no release found")?.to_string();
     let releases = format!("https://github.com/{repo}/releases");
     let html_url = v["html_url"].as_str().and_then(|u| safe_release_html_url(repo, u)).unwrap_or_else(|| releases.clone());
     let prefix = download_prefix(repo);
@@ -284,7 +284,18 @@ pub fn release_from_sums(repo: &str, tag: &str, sums: &[(String, String)]) -> Re
 /// The tag in a `https://github.com/<repo>/releases/tag/<tag>` redirect target.
 pub fn tag_from_redirect(location: &str, repo: &str) -> Option<String> {
     let rest = location.strip_prefix(&format!("https://github.com/{repo}/releases/tag/"))?;
-    (!rest.is_empty() && rest.len() <= 64 && rest.bytes().all(|b| b.is_ascii_alphanumeric() || b"._-+".contains(&b))).then(|| rest.to_string())
+    valid_tag(rest).then(|| rest.to_string())
+}
+
+/// A release tag we'd put in a URL, a file or the UI: up to 64 of `[A-Za-z0-9._+-]`, not
+/// starting with a dot, so no separators, control or bidi characters and no `.`/`..`.
+pub fn valid_tag(tag: &str) -> bool {
+    tag.len() <= 64 && !tag.starts_with('.') && plain_name(tag)
+}
+
+/// Non-empty and only `[A-Za-z0-9._+-]`, the characters every Craft release tag and asset uses.
+fn plain_name(s: &str) -> bool {
+    !s.is_empty() && s.bytes().all(|b| b.is_ascii_alphanumeric() || b"._-+".contains(&b))
 }
 
 /// Reads `sha256sum` output: `<hex>  <name>` (or `<hex> *<name>`) per line.
@@ -298,14 +309,16 @@ pub fn parse_sha256sums(text: &str) -> Vec<(String, String)> {
         .collect()
 }
 
-/// A bare file name we'd be willing to save: no paths, no control characters.
+/// A bare file name we'd be willing to save: only `[A-Za-z0-9._+-]`, no leading or trailing dot,
+/// and not a name Windows reserves for a device (`CON`, `nul.txt`, `COM1.msi`).
 pub fn valid_file_name(name: &str) -> bool {
-    !name.is_empty()
-        && name.len() <= 200
-        && name != "."
-        && name != ".."
-        && !name.starts_with('.')
-        && !name.chars().any(|c| c.is_control() || matches!(c, '/' | '\\' | ':' | '<' | '>' | '"' | '|' | '?' | '*'))
+    name.len() <= 200 && !name.starts_with('.') && !name.ends_with('.') && plain_name(name) && !windows_device(name)
+}
+
+fn windows_device(name: &str) -> bool {
+    let stem = name.split('.').next().unwrap_or_default().to_ascii_uppercase();
+    matches!(stem.as_str(), "CON" | "PRN" | "AUX" | "NUL")
+        || matches!(stem.as_bytes(), [b'C', b'O', b'M', b'1'..=b'9'] | [b'L', b'P', b'T', b'1'..=b'9'])
 }
 
 /// `s` cut to at most `max` bytes (on a char boundary) with an ellipsis, for text from a
@@ -524,9 +537,56 @@ mod tests {
 
     #[test]
     fn file_names_cannot_escape() {
-        for bad in ["", "..", "a/b", "a\\b", ".hidden", "c:x", "x\u{0}"] {
+        for bad in [
+            "",
+            "..",
+            "a/b",
+            "a\\b",
+            ".hidden",
+            "c:x",
+            "x\u{0}",
+            "a b.dmg",
+            "a%2e.msi",
+            "a#b",
+            "evil\u{202e}gpj.exe",
+            "x.msi.",
+            "CON",
+            "nul.txt",
+            "COM1.msi",
+            "lpt9",
+            "prn",
+            "aux.log",
+            "a".repeat(201).as_str(),
+        ] {
             assert!(!valid_file_name(bad), "{bad:?}");
         }
-        assert!(valid_file_name("photocraft-0.5.0-linux-x86_64.AppImage"));
+        for good in [
+            "photocraft-0.5.0-linux-x86_64.AppImage",
+            "lightcraft_0.9.0_amd64.deb",
+            "x-1.0+git.tar.gz",
+            "console.zip",
+            "COM10.txt",
+            "COM0.msi",
+            "lpt0",
+        ] {
+            assert!(valid_file_name(good), "{good:?}");
+        }
+    }
+
+    #[test]
+    fn tags_are_plain() {
+        for good in ["v0.6.0", "v1.0.0-beta.2", "nightly-2026-10-29", "v1+build.7"] {
+            assert!(valid_tag(good), "{good:?}");
+        }
+        for bad in ["", ".", "..", ".v1", "v1/x", "v1\\x", "v1\nx", "v1\u{0}x", "v1\u{202e}", "v 1", "v1?x", "v".repeat(65).as_str()] {
+            assert!(!valid_tag(bad), "{bad:?}");
+        }
+        let api = r#"{"tag_name": "v1\u202egpj", "html_url": "", "assets": []}"#;
+        assert!(parse_github_release(api, REPO).is_err());
+        let prefix = download_prefix(REPO);
+        assert!(download_url_under(&format!("{prefix}v1/a.dmg"), &prefix));
+        assert!(!download_url_under(&format!("{prefix}v1\u{202e}/a.dmg"), &prefix), "an odd tag in a download link");
+        assert!(!download_url_under(&format!("{prefix}v1/a\u{202e}.dmg"), &prefix), "an odd name in a download link");
+        assert_eq!(tag_from_redirect(&format!("https://github.com/{REPO}/releases/tag/v1\u{202e}"), REPO), None);
     }
 }