From 72012aea855ec0035c573f1da7b73cf27f8221df Mon Sep 17 00:00:00 2001
From: CleverThis <hal9000@cleverthis.com>
Date: Sun, 17 May 2026 13:49:20 +0000
Subject: [PATCH] SEC: Fix LSP header injection via strict ASCII decoding

Replace errors='replace' with strict ascii.decode() in transport.py and
server.py header parsing to prevent non-ASCII byte injection attacks.
The replacement character (U+FFFD) could be exploited to craft malformed
headers that bypass validation yet alter message framing per the LSP
specification's ASCII-only header requirement. Closes #10608.
---
 src/cleveragents/lsp/server.py    | 22 +++++++++++++++++++---
 src/cleveragents/lsp/transport.py | 10 +++++++++-
 2 files changed, 28 insertions(+), 4 deletions(-)

diff --git a/src/cleveragents/lsp/server.py b/src/cleveragents/lsp/server.py
index a3a3c363d..e5679a915 100644
--- a/src/cleveragents/lsp/server.py
+++ b/src/cleveragents/lsp/server.py
@@ -169,13 +169,22 @@ class LspServer:
             raw_line = self._input.readline(MAX_HEADER_LINE_LENGTH)
             if not raw_line:
                 return
+            # SEC: Strict ASCII decoding prevents header injection attacks.
+            # Using ``errors="replace"`` silently accepts non-ASCII bytes as
+            # the replacement character (U+FFFD), which can mask malformed
+            # headers and cause transport desync.  The LSP spec requires
+            # all HTTP-style headers to be plain ASCII.  On decode errors,
+            # ``continue`` keeps draining remaining header lines so that
+            # the stream is correctly positioned at the blank terminator.
             try:
                 decoded_line = raw_line.decode("ascii").rstrip("\r\n")
-            except UnicodeDecodeError:
+            except UnicodeDecodeError as exc:
                 logger.warning(
                     "lsp.transport.non_ascii_header_byte",
+                    raw_line=repr(raw_line),
+                    error=str(exc),
                 )
-                return None
+                continue  # keep draining; line already consumed by readline()
             line = decoded_line
             if line == "":
                 return
@@ -231,11 +240,18 @@ class LspServer:
                 )
                 self._drain_headers()
                 return _SKIP
+            # SEC: Strict ASCII decoding prevents header injection attacks.
+            # Using ``errors="replace"`` silently accepts non-ASCII bytes as
+            # the replacement character (U+FFFD), which can mask malformed
+            # headers and cause transport desync.  The LSP spec requires
+            # all HTTP-style headers to be plain ASCII.
             try:
                 decoded_line = raw_line.decode("ascii").rstrip("\r\n")
-            except UnicodeDecodeError:
+            except UnicodeDecodeError as exc:
                 logger.warning(
                     "lsp.transport.non_ascii_header_byte",
+                    raw_line=repr(raw_line),
+                    error=str(exc),
                 )
                 return _SKIP
             line = decoded_line
diff --git a/src/cleveragents/lsp/transport.py b/src/cleveragents/lsp/transport.py
index b9f903aa4..fedd55135 100644
--- a/src/cleveragents/lsp/transport.py
+++ b/src/cleveragents/lsp/transport.py
@@ -254,11 +254,19 @@ class StdioTransport:
             line = stdout.readline()
             if not line:
                 return None  # EOF — server exited
+            # SEC: Strict ASCII decoding prevents header injection attacks.
+            # Using ``errors="replace"`` silently accepts non-ASCII bytes as
+            # the replacement character (U+FFFD), which an attacker could use
+            # to craft malformed headers that bypass validation yet alter
+            # message framing.  The LSP specification requires all HTTP-style
+            # headers to be plain ASCII; reject anything else immediately.
             try:
                 decoded = line.decode("ascii").strip()
-            except UnicodeDecodeError:
+            except UnicodeDecodeError as exc:
                 logger.warning(
                     "lsp.transport.non_ascii_header_byte",
+                    raw_line=repr(line),
+                    error=str(exc),
                 )
                 return None
             if not decoded: