From bd4cf9fb7fc6d5f0979958c01825228f8db9b7d3 Mon Sep 17 00:00:00 2001
From: "K. Isom" <kyle@imap.cc>
Date: Mon, 28 Sep 2026 14:12:12 -0700
Subject: [PATCH] review: a key split by an invisible character is still a
 secret

has_secret masked each character a character rule matches to a space, so one
zero-width space inside a key cut it into two runs too short to match. The
secret rules now also run on the text with those characters deleted, and the
text is a secret if either matches.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
---
 crates/memoryd/src/honcho/decide.rs          | 29 +++++++------
 crates/memoryd/tests/honcho_decide_hidden.rs | 44 ++++++++++++++++++++
 2 files changed, 60 insertions(+), 13 deletions(-)
 create mode 100644 crates/memoryd/tests/honcho_decide_hidden.rs

diff --git a/crates/memoryd/src/honcho/decide.rs b/crates/memoryd/src/honcho/decide.rs
index 5c15b45..62e493c 100644
--- a/crates/memoryd/src/honcho/decide.rs
+++ b/crates/memoryd/src/honcho/decide.rs
@@ -78,21 +78,24 @@ pub fn peer_name(speaker: &str) -> String {
 
 /// Whether `text` holds a known-prefix secret, by `refuse`'s secret rules alone. `refuse` gives
 /// only the first rule a text breaks, and its character rules come first, so each character they
-/// match is first replaced by a space: it is not a key character and not a letter, so a secret's
-/// prefix and run are found as before, and a carriage return cannot hide a key.
+/// match is taken out of the way, in two ways, and the text is a secret if either is: replaced by
+/// a space, which is not a key character and not a letter, so a secret's prefix and run are found
+/// as before and a carriage return cannot hide a key; and deleted, so that one inside a key cannot
+/// split it into two runs too short to match.
 pub fn has_secret(text: &str) -> bool {
     let mut buf = [0u8; 4];
-    let masked: String = text
-        .chars()
-        .map(|c| {
-            if refuse(c.encode_utf8(&mut buf)).is_some() {
-                ' '
-            } else {
-                c
-            }
-        })
-        .collect();
-    refuse(&masked).is_some_and(|r| r.rule.starts_with("secret-"))
+    let mut masked = String::with_capacity(text.len());
+    let mut deleted = String::with_capacity(text.len());
+    for c in text.chars() {
+        if refuse(c.encode_utf8(&mut buf)).is_some() {
+            masked.push(' ');
+        } else {
+            masked.push(c);
+            deleted.push(c);
+        }
+    }
+    let secret = |t: &str| refuse(t).is_some_and(|r| r.rule.starts_with("secret-"));
+    secret(&masked) || secret(&deleted)
 }
 
 /// The turn's user text and its speaker, from its first record, the `User` one.
diff --git a/crates/memoryd/tests/honcho_decide_hidden.rs b/crates/memoryd/tests/honcho_decide_hidden.rs
new file mode 100644
index 0000000..7e2c28c
--- /dev/null
+++ b/crates/memoryd/tests/honcho_decide_hidden.rs
@@ -0,0 +1,44 @@
+//! A key split by an invisible character is still a secret (review of M5c-1, finding 5): masking
+//! the character to a space would cut the key into two short runs, so the text with it deleted is
+//! checked too. Every key here is built from pieces so that no file holds one whole.
+
+use memoryd::honcho::decide::has_secret;
+
+fn mixed(n: usize) -> String {
+    "A1b2".chars().cycle().take(n).collect()
+}
+
+#[test]
+fn an_invisible_character_inside_a_key_does_not_hide_it() {
+    let github = ["gh", "p_"].concat();
+    let sk = ["sk", "-proj-"].concat();
+    let aws = ["AK", "IA"].concat();
+    for mark in ["\u{200b}", "\u{7}", "\r", "\u{e000}", "\u{fdd0}", "\u{ad}"] {
+        let texts = [
+            format!("{github}{}{mark}{}", mixed(18), mixed(18)),
+            format!("{github}{mark}{}", mixed(36)),
+            format!("{}{mark}{}{}", &github[..2], &github[2..], mixed(36)),
+            format!("{sk}{}{mark}{}", mixed(20), mixed(20)),
+            format!("{aws}IOSF{mark}ODNN7EXAMPLE"),
+            format!(
+                "my token is {github}{}{mark}{}, keep it",
+                mixed(30),
+                mixed(6)
+            ),
+        ];
+        for text in &texts {
+            assert!(has_secret(text), "{text:?}");
+        }
+    }
+}
+
+#[test]
+fn deleting_invisible_characters_finds_no_secret_in_plain_text() {
+    for text in [
+        "a zero width\u{200b}space",
+        "gh\u{200b}p_ is how GitHub tokens start",
+        "soft\u{ad}hyphen and sk-learn",
+    ] {
+        assert!(!has_secret(text), "{text:?}");
+    }
+}