diff --git a/crates/README.md b/crates/README.md
new file mode 100644
index 00000000000..75afd1cf111
--- /dev/null
+++ b/crates/README.md
@@ -0,0 +1,10 @@
+# crates
+
+This directory contains the crates that make up Edit and its supporting tooling.
+
+* `edit`: Main editor binary and library<br>
+  It is split apart into a library to allow for benchmarks.
+* `lsh`: Syntax-highlighting compiler and runtime
+* `lsh-bin`: A small CLI for experimenting with and debugging LSH output
+* `stdext`: Shared utility code used across the workspace
+* `unicode-gen`: Code generation utilities for Unicode LUTs
diff --git a/crates/lsh-bin/src/main.rs b/crates/lsh-bin/src/main.rs
index 365212bbbb8..68547c431d5 100644
--- a/crates/lsh-bin/src/main.rs
+++ b/crates/lsh-bin/src/main.rs
@@ -31,24 +31,24 @@ enum SubCommands {
 #[derive(FromArgs, PartialEq, Debug)]
 #[argh(subcommand, name = "compile", description = "Generate Rust code from .lsh files")]
 struct SubCommandOneCompile {
-    #[argh(positional, description = "source .lsh file or directory")]
-    lsh: PathBuf,
+    #[argh(positional, description = "source .lsh files or directories")]
+    lsh: Vec<PathBuf>,
 }
 
 #[derive(FromArgs, PartialEq, Debug)]
 #[argh(subcommand, name = "assembly", description = "Generate assembly from .lsh files")]
 struct SubCommandAssembly {
-    #[argh(positional, description = "source .lsh file or directory")]
-    lsh: PathBuf,
+    #[argh(positional, description = "source .lsh files or directories")]
+    lsh: Vec<PathBuf>,
 }
 
 #[derive(FromArgs, PartialEq, Debug)]
 #[argh(subcommand, name = "render", description = "Highlight text files")]
 struct SubCommandRender {
-    #[argh(positional, description = "source .lsh file or directory")]
-    lsh: PathBuf,
-    #[argh(positional, description = "source text file")]
+    #[argh(option, description = "source text file")]
     input: PathBuf,
+    #[argh(positional, description = "source .lsh files or directories")]
+    lsh: Vec<PathBuf>,
 }
 
 pub fn main() {
@@ -67,21 +67,32 @@ fn run() -> anyhow::Result<()> {
     let mut read_lsh = |path: &Path| {
         if path.is_dir() { generator.read_directory(path) } else { generator.read_file(path) }
     };
+    let mut read_lsh_inputs = |paths: &[PathBuf]| -> anyhow::Result<()> {
+        if paths.is_empty() {
+            bail!("At least one .lsh file or directory is required");
+        }
+
+        for path in paths {
+            read_lsh(path)?;
+        }
+
+        Ok(())
+    };
 
     match &command.sub {
         SubCommands::Compile(cmd) => {
-            read_lsh(&cmd.lsh)?;
+            read_lsh_inputs(&cmd.lsh)?;
             let output = generator.generate_rust()?;
             _ = stdout().write_all(output.as_bytes());
         }
         SubCommands::Assembly(cmd) => {
-            read_lsh(&cmd.lsh)?;
+            read_lsh_inputs(&cmd.lsh)?;
             let vt = stdout().is_terminal();
             let output = generator.generate_assembly(vt)?;
             _ = stdout().write_all(output.as_bytes());
         }
         SubCommands::Render(cmd) => {
-            read_lsh(&cmd.lsh)?;
+            read_lsh_inputs(&cmd.lsh)?;
             run_render(generator, &cmd.input)?;
         }
     }
diff --git a/crates/lsh/README.md b/crates/lsh/README.md
new file mode 100644
index 00000000000..bf51865266e
--- /dev/null
+++ b/crates/lsh/README.md
@@ -0,0 +1,26 @@
+# lsh
+
+`lsh` contains the compiler and runtime for Edit's syntax-highlighting system.
+
+At a high level:
+* Language definitions live in `definitions/*.lsh`
+* The compiler lowers them into bytecode
+* The runtime executes the bytecode on the input text line by line
+
+To understand the definition language itself, read [definitions/README.md](definitions/README.md).
+
+For debugging and optimizing language definitions use `lsh-bin`.
+To see the generated assembly, for example:
+```sh
+# Show the generated assembly of a file or directory
+cargo run -p lsh-bin -- assembly crates/lsh/definitions/diff.lsh
+
+# Due to the lack of include statements, you must specify included files manually.
+# Here, git_commit.lsh implicitly relies on diff() from diff.lsh.
+cargo run -p lsh-bin -- assembly crates/lsh/definitions/git_commit.lsh crates/lsh/definitions/diff.lsh
+```
+
+Or to render a file:
+```sh
+cargo run -p lsh-bin -- render --input assets/highlighting-tests/html.html crates/lsh/definitions
+```
diff --git a/crates/lsh/definitions/README.md b/crates/lsh/definitions/README.md
new file mode 100644
index 00000000000..b0a29f609f3
--- /dev/null
+++ b/crates/lsh/definitions/README.md
@@ -0,0 +1,153 @@
+# LSH Definitions
+
+This directory contains syntax highlighting definitions.
+Each `.lsh` file describes how to highlight one or more file types.
+The compiler turns these definitions into bytecode, and the runtime executes that bytecode against the input one line at a time.
+
+Essentially, LSH is a small, line-oriented coroutine language for writing lexers.
+
+## The basic idea
+
+Most definitions follow the same pattern:
+* Select a definition by file name or path
+* Walk the current line from left to right
+* Try regexes at the current position
+* `yield` highlight kinds as tokens are recognized
+* Use `await input` only when a construct needs to continue onto the next line
+
+## A minimal definition
+
+A definition is a `pub fn` with attributes that tell the editor when to use it:
+
+```rs
+#[display_name = "Diff"]
+#[path = "**/*.diff"]
+#[path = "**/*.patch"]
+pub fn diff() {
+    if /(?:diff|---|\+\+\+).*/ {
+        yield meta.header;
+    } else if /-.*/ {
+        yield markup.deleted;
+    } else if /\+.*/ {
+        yield markup.inserted;
+    }
+}
+```
+
+`#[display_name]` sets the human-readable name.
+`#[path]` is a glob pattern; you can have as many as you need.
+Functions without `pub` are private helpers that can be called from other definitions.
+
+## How execution works
+
+The runtime feeds input to a definition one line at a time.
+Within a line, matching is always left to right.
+
+Each `if /regex/` tries to match at the current position:
+* On success, the input position advances past the match and the block runs
+* On failure, the input position does not move and the `else` branch, if any, runs
+
+Definitions behave like coroutines:
+* If execution reaches `await input`, the function suspends and resumes on the next line
+* If the function returns, the next line starts again from the top of the function
+
+## Highlighting with `yield`
+
+`yield <kind>` emits a highlight span.
+Everything between the previous `yield` and the current position is colored with `<kind>`.
+
+> [!NOTE]
+> This can be confusing in practice, because `yield` does not just color the regex it appears in.
+> Long term, the goal is for `yield` to apply only to the regex it appears in, or to some other explicitly specified range.
+
+Highlight kinds are dotted identifiers such as `comment`, `string`, `keyword.control`, `constant.numeric`, and `markup.bold`.
+Kinds are interned at compile time. You can invent new ones, but the editor still needs to know what color to assign them.
+
+`yield other` switches back to the default, unhighlighted kind.
+Use it when you want to reset the current highlight between tokens. See [json.lsh](json.lsh) for a representative pattern.
+
+## Multi-line constructs
+
+Single-line constructs need no special handling.
+For constructs that can span lines, such as block comments or fenced code blocks, combine `loop` or `until` with `await input`:
+
+```rs
+if /\/\*/ {
+    loop {
+        yield comment;
+        await input;
+        if /\*\// {
+            yield comment;
+            break;
+        }
+    }
+}
+```
+
+`await input` means "advance to the next line if there is no more input to consume here."
+If there is still unconsumed text on the current line, it is a no-op and execution continues immediately.
+
+One important detail: if you want the remainder of the current line to stay highlighted, emit the appropriate `yield` before `await input`.
+
+## Control flow
+
+| Expression | Meaning |
+|------------|---------|
+| `if /pat/ { ... }` | Match `pat` at the current position and enter the block on success |
+| `else if /pat/ { ... }` | Try another pattern if the previous one failed |
+| `else { ... }` | Fallback branch |
+| `loop { ... }` | Loop until `break`, `continue`, or `return` |
+| `until /pat/ { ... }` | Repeat the body until `pat` matches, then consume the match and exit |
+| `break` | Exit the innermost loop |
+| `continue` | Restart the innermost loop |
+| `return` | Exit the current function |
+
+`until /$/ { ... }` is the usual way to say "keep processing until end-of-line."
+
+## Capture groups
+
+Regexes can have capture groups.
+Use `yield $N as <kind>` when only part of the match should receive a specific highlight:
+
+```rs
+if /([\w:.-]+)\s*=/ {
+    yield $1 as variable;
+    yield other;
+}
+```
+
+The full regex match is still consumed.
+Only capture group `$1` receives the `variable` highlight; everything else falls through to the following `yield`.
+
+## Variables and the input position
+
+You can store the current input offset in a variable and compare against it later:
+
+```rs
+var indentation = off;
+// ...later...
+if off <= indentation {
+    break;
+}
+```
+
+`off` is the built-in register for the current position in the line.
+[yaml.lsh](yaml.lsh) uses this pattern to detect when a multi-line string ends.
+
+## Calling other definitions
+
+Definitions can call helper functions or other definitions.
+This is how [markdown.lsh](markdown.lsh) delegates the contents of fenced code blocks:
+
+```rs
+if /(?i:json)/ {
+    loop {
+        await input;
+        if /\s*```/ { return; }
+        else { json(); if /.*/ {} }
+    }
+}
+```
+
+The `if /.*/ {}` at the end consumes any text that the nested definition did not consume itself.
+Without that final match, `await input` would see remaining input on the current line and continue immediately instead of advancing to the next line.
diff --git a/crates/lsh/src/lib.rs b/crates/lsh/src/lib.rs
index 8f914e202ca..328fbcd1e3e 100644
--- a/crates/lsh/src/lib.rs
+++ b/crates/lsh/src/lib.rs
@@ -1,7 +1,7 @@
 // Copyright (c) Microsoft Corporation.
 // Licensed under the MIT License.
 
-//! Welcome to Leonard's Syntax Highlighter (LSH), otherwise known as
+//! Welcome to the Lightweight Syntax Highlighter (LSH), otherwise known as
 //! Leonard's Shitty Highlighter, which is really what it is.
 //!
 //! ## Architecture