fix(routing): pin managed skill payloads to LF (#502)

.gitattributes

@@ -0,0 +1,3 @@
+# These files are embedded verbatim by include_str!, so checkout EOLs become
+# part of the CLI and MCP payload contract.
+crates/ai-memory-core/src/routing_skills/*/SKILL.md text eol=lf

CHANGELOG.md

@@ -10,6 +10,11 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## [1.32.2] - 2026-08-26
 
 ### Fixed
+- Made managed routing `SKILL.md` payloads byte-identical across release
+  platforms. Windows builds previously embedded CRLF from the runner checkout
+  while Linux and macOS builds embedded LF, so one tag returned different
+  bytes through CLI installs and `memory_install_self_routing`. The embedded
+  assets now use LF everywhere without rewriting user-authored files. (#502)
 - The native client now trusts CAs from the platform trust store. It was built
   with reqwest's `rustls-tls`, which bundles the Mozilla webpki roots and
   ignores the OS store, so a CA the operator had installed locally — Caddy's

README.md

@@ -629,6 +629,8 @@ one matching entry.
   (`ai-memory install-instructions`, or `--target AGENTS.md` for AGENTS-based
   projects) when you want new tool guidance. The refresh writes the slim
   markered snippet and managed Agent Skills from the same binary-owned assets.
+  Managed skill payloads use LF line endings on every release platform, while
+  user-authored files keep their existing line endings.
 
 For every client in the [Support Matrix](#support-matrix), plus curl-based hook
 installs, source builds, CLI environment variables, and the full subcommand

crates/ai-memory-core/src/routing_skills.rs

@@ -146,6 +146,22 @@ mod tests {
         }
     }
 
+    #[test]
+    fn managed_skill_payloads_use_lf_line_endings() {
+        for skill in MANAGED_SKILLS {
+            assert!(
+                !skill.content.contains('\r'),
+                "{} must use LF line endings",
+                skill.name
+            );
+            assert!(
+                skill.content.ends_with('\n'),
+                "{} must end with a newline",
+                skill.name
+            );
+        }
+    }
+
     #[test]
     fn relative_paths_are_safe_relative_skill_markdown_files() {
         for skill in MANAGED_SKILLS {
@@ -197,8 +213,7 @@ mod tests {
     }
 
     fn parse_frontmatter(skill: &ManagedSkill) -> Frontmatter {
-        let content = skill.content.replace("\r\n", "\n");
-        let Some(rest) = content.strip_prefix("---\n") else {
+        let Some(rest) = skill.content.strip_prefix("---\n") else {
             panic!("{} must start with frontmatter", skill.name);
         };
         let Some((frontmatter, _body)) = rest.split_once("\n---\n") else {

docs/install.md

@@ -1752,6 +1752,9 @@ with the slim snippet, leaves unrelated instructions before and after it alone,
 and writes a timestamped `.bak-*` backup before changing an existing file.
 Managed skill files contain an ai-memory ownership marker; same-name user skills
 without that marker are preserved unless you explicitly force replacement.
+Their embedded payloads use LF line endings on every release platform, so CLI
+installs and `memory_install_self_routing` return the same bytes on Windows,
+Linux, and macOS. This does not rewrite line endings in user-authored files.
 `install-instructions --print` previews only the instruction snippet; run
 `install-skills --print` when you want to preview the managed skill payloads.