diff --git a/UnitTests/pch.hpp b/UnitTests/pch.hpp
index 83b331acb..6188944f0 100644
--- a/UnitTests/pch.hpp
+++ b/UnitTests/pch.hpp
@@ -16,6 +16,8 @@
 #include <stdex/locale.hpp>
 #include <stdex/mapping.hpp>
 #include <stdex/math.hpp>
+#include <stdex/memory.hpp>
+#include <stdex/minisign.hpp>
 #include <stdex/parser.hpp>
 #include <stdex/pool.hpp>
 #include <stdex/progress.hpp>
diff --git a/include/stdex/minisign.hpp b/include/stdex/minisign.hpp
new file mode 100644
index 000000000..326ca3ba9
--- /dev/null
+++ b/include/stdex/minisign.hpp
@@ -0,0 +1,126 @@
+/*
+	SPDX-License-Identifier: MIT
+	Copyright © 2023-2024 Amebis
+*/
+
+#pragma once
+
+#include "base64.hpp"
+#include "compat.hpp"
+#include "parser.hpp"
+#include "stream.hpp"
+#include <stdint.h>
+#include <string>
+#include <vector>
+
+namespace stdex
+{
+	namespace minisign
+	{
+		///
+		/// Test for "untrusted comment:"
+		///
+		class untrusted_comment : public stdex::parser::parser
+		{
+		protected:
+			virtual bool do_match(
+				_In_reads_or_z_opt_(end) const char* text,
+				_In_ size_t start = 0,
+				_In_ size_t end = SIZE_MAX,
+				_In_ int flags = stdex::parser::match_default)
+			{
+				_Unreferenced_(flags);
+				_Assume_(text || start + 17 >= end);
+				if (start + 17 < end &&
+					text[start + 0] == 'u' &&
+					text[start + 1] == 'n' &&
+					text[start + 2] == 't' &&
+					text[start + 3] == 'r' &&
+					text[start + 4] == 'u' &&
+					text[start + 5] == 's' &&
+					text[start + 6] == 't' &&
+					text[start + 7] == 'e' &&
+					text[start + 8] == 'd' &&
+					text[start + 9] == ' ' &&
+					text[start + 10] == 'c' &&
+					text[start + 11] == 'o' &&
+					text[start + 12] == 'm' &&
+					text[start + 13] == 'm' &&
+					text[start + 14] == 'e' &&
+					text[start + 15] == 'n' &&
+					text[start + 16] == 't' &&
+					text[start + 17] == ':')
+				{
+					this->interval.end = (this->interval.start = start) + 18;
+					return true;
+				}
+				this->interval.invalidate();
+				return false;
+			}
+		};
+
+		///
+		/// Test for CRLF or LF
+		///
+		class line_break : public stdex::parser::parser
+		{
+		protected:
+			virtual bool do_match(
+				_In_reads_or_z_opt_(end) const char* text,
+				_In_ size_t start = 0,
+				_In_ size_t end = SIZE_MAX,
+				_In_ int flags = stdex::parser::match_default)
+			{
+				_Unreferenced_(flags);
+				_Assume_(text || start + 1 >= end);
+				if (start + 1 < end &&
+					text[start + 0] == '\r' &&
+					text[start + 1] == '\n')
+				{
+					this->interval.end = (this->interval.start = start) + 2;
+					return true;
+				}
+				_Assume_(text || start >= end);
+				if (start < end && text[start] == '\n') {
+					this->interval.end = (this->interval.start = start) + 1;
+					return true;
+				}
+				this->interval.invalidate();
+				return false;
+			}
+		};
+
+		///
+		/// Parses .minisig file
+		///
+		/// \param[in,out] minisig    Stream with position set to the beginning of the Minisign signature. Typically a .minisig file.
+		/// \param[out]    algorithm  Minisign algorithm used to create signature: 'd' legacy, 'D' hashed
+		/// \param[out]    key_id     8 random bytes, matching the public key used to sign content
+		/// \param[out]    signature  ed25519(<file data>) when using legacy algorithm; ed25519(Blake2b-512(<file data>)) when using hashed algorithm.
+		///
+		inline void parse_minisig(_Inout_ stdex::stream::basic& minisig, _Out_ uint8_t& algorithm, _Out_writes_all_(8) uint8_t key_id[8], _Out_writes_all_(32) uint8_t signature[64])
+		{
+			std::vector<uint8_t> data;
+			minisign::untrusted_comment untrusted_comment;
+			std::string line;
+			for (;;) {
+				minisig.readln(line);
+				if (!minisig.ok())
+					break;
+				if (line.empty() ||
+					untrusted_comment.match(line.data(), 0, line.size()))
+					continue;
+				stdex::base64_dec decoder; bool is_last;
+				decoder.decode(data, is_last, line.data(), line.size());
+				break;
+			}
+			if (data.size() < 74)
+				throw std::runtime_error("Minisign signature is too short");
+			if (data[0] != 'E')
+				throw std::runtime_error("not a Minisign signature");
+			algorithm = data[1];
+			memcpy(&key_id[0], &data[2], 8);
+			memcpy(&signature[0], &data[10], 64);
+		}
+	}
+}
\ No newline at end of file