diff --git a/CHANGELOG.md b/CHANGELOG.md
index 8d3b64a..87d4526 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -4,6 +4,7 @@ All notable changes to this project will be documented in this file.
 
 ## Unreleased
 
+- **Binary files are reported, not dumped; the header no longer leaks absolute paths (#21 P11).** A file with NUL bytes in its first 8 KiB (UTF-16/32 text with a BOM excepted) is listed under `## Skipped` as `binary` instead of a `<Binary file or unsupported encoding>` block in the body (a repo without `.git` produced hundreds of those). The header says `from the `my proj` directory` rather than the absolute path passed with `-d`.
 - **`--stats[=N]` per-file token breakdown; real token count after a run (#21 P7).** `--stats` lists the largest files by tokens (20 by default; `--stats=0` lists all) and, with `--max-tokens`, which files fit, which is cut first and what is omitted; it writes no document. The `Estimated tokens` line after a run is now the tokenizer's count of the written document (it was bytes/4, off by −11% to −75% on real code); output piped with `-o -` keeps the `~NK` estimate.
 - **Auto-diff honors `--max-tokens` like the standard renderer (#20 B14).** The auto-diff path used a bytes/4 cut over the finished document (mid-file, ignoring `--encoding`, omissions unreported). Files are now budgeted one by one with the real tokenizer: a file that would cross the budget is clamped at a `--truncate` boundary (and says so) or omitted, and the notice counts the omitted files. The budget search is shared (`markdown::fit_chunk_to_budget`) with the standard path.
 - **`--structure` lists imports and exports (#22 B15).** The help promised them but only counts were printed (and the internal lists held placeholders). Each file now gets `**Imports:**` / `**Exports:**` lists (Rust `use`/`pub use`, Python, JS/TS `import`/`export`, Go, Java, C/C++ `#include`), capped at 20 entries. Top-level declarations only.
diff --git a/src/content_filter.rs b/src/content_filter.rs
index 4325225..dde52ee 100644
--- a/src/content_filter.rs
+++ b/src/content_filter.rs
@@ -222,6 +222,10 @@ pub enum SkipReason {
     TooLarge,
     /// Filename (or `.npmrc` / `.pypirc` contents) looks like a credential.
     Secret,
+    /// Not a known asset type, but the content is binary (NUL bytes in the
+    /// first 8 KiB, outside UTF-16/32 text). Rendering it would only print a
+    /// placeholder block.
+    Binary,
 }
 
 impl SkipReason {
@@ -231,6 +235,7 @@ impl SkipReason {
             SkipReason::Asset => "asset",
             SkipReason::TooLarge => "too large",
             SkipReason::Secret => "secret",
+            SkipReason::Binary => "binary",
         }
     }
 }
@@ -395,9 +400,26 @@ pub fn classify(
     {
         return Some((SkipReason::TooLarge, "too large"));
     }
+    if looks_binary(path) {
+        return Some((SkipReason::Binary, "binary"));
+    }
     None
 }
 
+/// Sniff the start of the file. Unreadable files are not reported as binary
+/// (the renderer deals with them).
+fn looks_binary(path: &Path) -> bool {
+    use std::io::Read;
+    let Ok(file) = std::fs::File::open(path) else {
+        return false;
+    };
+    let mut head = Vec::with_capacity(8192);
+    if file.take(8192).read_to_end(&mut head).is_err() {
+        return false;
+    }
+    crate::markdown::is_binary_content(&head)
+}
+
 /// Stderr lines for a skip list: one warning per secret (path and category
 /// only), then a single summary line. Empty when nothing was skipped.
 pub fn report_lines(skipped: &[SkippedFile]) -> Vec<String> {
@@ -453,11 +475,13 @@ pub fn summary_line(skipped: &[SkippedFile]) -> Option<String> {
     let mut assets = 0usize;
     let mut large = 0usize;
     let mut secrets = 0usize;
+    let mut binaries = 0usize;
     for item in skipped {
         match item.reason {
             SkipReason::Asset => assets += 1,
             SkipReason::TooLarge => large += 1,
             SkipReason::Secret => secrets += 1,
+            SkipReason::Binary => binaries += 1,
         }
     }
     let mut parts = Vec::new();
@@ -470,6 +494,9 @@ pub fn summary_line(skipped: &[SkippedFile]) -> Option<String> {
     if secrets > 0 {
         parts.push(count_phrase(secrets, "secret", "secrets"));
     }
+    if binaries > 0 {
+        parts.push(count_phrase(binaries, "binary", "binaries"));
+    }
     let noun = if skipped.len() == 1 { "file" } else { "files" };
     Some(format!(
         "Skipped {} {} ({}).",
diff --git a/src/lib.rs b/src/lib.rs
index 0d96457..f184766 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -423,7 +423,7 @@ pub fn run_with_args(args: Args, config: Config, prompter: &impl Prompter) -> io
                     encoding,
                     &format!(
                         "This document contains files from the {} directory with extensions: {} \n",
-                        fences::inline_code(&final_args.input),
+                        fences::inline_code(&markdown::directory_display_name(&final_args.input)),
                         final_args.filter.join(", ")
                     ),
                 );
@@ -432,7 +432,7 @@ pub fn run_with_args(args: Args, config: Config, prompter: &impl Prompter) -> io
                     encoding,
                     &format!(
                         "This document contains all files from the {} directory, optimized for LLM consumption.\n",
-                        fences::inline_code(&final_args.input)
+                        fences::inline_code(&markdown::directory_display_name(&final_args.input))
                     ),
                 );
             }
diff --git a/src/markdown.rs b/src/markdown.rs
index 063ec9f..a235329 100644
--- a/src/markdown.rs
+++ b/src/markdown.rs
@@ -113,16 +113,7 @@ pub fn generate_markdown(
         path: (output_path != "-").then(|| std::path::PathBuf::from(output_path)),
     };
 
-    let input_dir_name = if input_dir == "." {
-        let current_dir = std::env::current_dir()?;
-        current_dir
-            .file_name()
-            .and_then(|n| n.to_str())
-            .unwrap_or_else(|| current_dir.to_str().unwrap_or("project"))
-            .to_string()
-    } else {
-        input_dir.to_string()
-    };
+    let input_dir_name = directory_display_name(input_dir);
 
     // --- Header + file tree (buffered) --- //
     // Build the header and tree into a buffer first so we can (a) write them in
@@ -1109,6 +1100,26 @@ fn first_visibility_warning_for(ext: &str) -> bool {
     guard.insert(ext.to_string())
 }
 
+/// The name shown for the input directory in the document header: its last
+/// path component, never the absolute path (which leaks the user's home
+/// directory and username into a document that is meant to be shared).
+pub(crate) fn directory_display_name(input_dir: &str) -> String {
+    use std::path::Component;
+    let path = Path::new(input_dir);
+    // The lexical name the user typed wins (a symlink keeps its own name).
+    let last = path.components().next_back();
+    if let Some(Component::Normal(name)) = last
+        && let Some(name) = name.to_str()
+    {
+        return name.to_string();
+    }
+    // `.`, `..`, `/`, `C:\`: resolve, and never fall back to a path.
+    fs::canonicalize(path)
+        .ok()
+        .and_then(|p| p.file_name().and_then(|n| n.to_str().map(str::to_string)))
+        .unwrap_or_else(|| "project".to_string())
+}
+
 /// Bytes read from the start of a file when deciding text vs binary.
 const BINARY_SNIFF_LEN: usize = 8192;
 
diff --git a/tests/test_binary_and_header.rs b/tests/test_binary_and_header.rs
new file mode 100644
index 0000000..3f0ef00
--- /dev/null
+++ b/tests/test_binary_and_header.rs
@@ -0,0 +1,174 @@
+//! Binary files are reported, not dumped as placeholder blocks, and the header
+//! names the directory instead of leaking an absolute path (#21 P11).
+
+use context_builder::cli::Args;
+use context_builder::config::Config;
+use context_builder::content_filter::{ContentPolicy, SkipReason, classify};
+use context_builder::{Prompter, run_with_args};
+use std::fs;
+use std::path::Path;
+use std::process::Command;
+
+struct Yes;
+
+impl Prompter for Yes {
+    fn confirm_overwrite(&self, _file_path: &str) -> std::io::Result<bool> {
+        Ok(true)
+    }
+}
+
+fn args(input: &Path, output: &Path) -> Args {
+    Args {
+        input: input.to_string_lossy().to_string(),
+        output: output.to_string_lossy().to_string(),
+        filter: vec![],
+        ignore: vec![],
+        line_numbers: false,
+        preview: false,
+        token_count: false,
+        yes: true,
+        diff_only: false,
+        clear_cache: false,
+        encoding: "o200k_base".to_string(),
+        init: false,
+        max_tokens: None,
+        signatures: false,
+        structure: false,
+        truncate: "smart".to_string(),
+        visibility: "all".to_string(),
+        file_metadata: false,
+        max_file_size: "256K".to_string(),
+        hidden: false,
+        include_secrets: false,
+        include_lockfiles: false,
+        stats: None,
+    }
+}
+
+fn project() -> tempfile::TempDir {
+    let dir = tempfile::Builder::new()
+        .prefix("my proj ")
+        .tempdir()
+        .unwrap();
+    fs::write(dir.path().join("a.txt"), "hello\n").unwrap();
+    // Not a known asset extension, but clearly binary.
+    fs::write(dir.path().join("blob.dat"), [0u8, 1, 2, 3, 0, 0, 255, 254]).unwrap();
+    // NUL-padded UTF-16 with a BOM is text.
+    let mut utf16 = vec![0xFF, 0xFE];
+    for c in "wide text\n".encode_utf16() {
+        utf16.extend_from_slice(&c.to_le_bytes());
+    }
+    fs::write(dir.path().join("wide.txt"), utf16).unwrap();
+    dir
+}
+
+#[test]
+fn classify_marks_nul_content_as_binary_but_not_utf16_or_plain_text() {
+    let dir = project();
+    let policy = ContentPolicy::new("256K", &[], false);
+    let verdict = |name: &str| {
+        let path = dir.path().join(name);
+        classify(&path, fs::metadata(&path).unwrap().len(), &policy).map(|(r, _)| r)
+    };
+    assert_eq!(verdict("blob.dat"), Some(SkipReason::Binary));
+    assert_eq!(verdict("a.txt"), None);
+    assert_eq!(verdict("wide.txt"), None);
+    // A path that cannot be opened is left to the renderer.
+    assert_eq!(classify(&dir.path().join("missing.dat"), 10, &policy), None);
+}
+
+#[test]
+fn document_skips_binary_in_the_body_and_lists_it() {
+    let dir = project();
+    let out = dir.path().join("out.md");
+    run_with_args(args(dir.path(), &out), Config::default(), &Yes).unwrap();
+    let doc = fs::read_to_string(&out).unwrap();
+    assert!(doc.contains("- `blob.dat` — binary"), "{doc}");
+    assert!(!doc.contains("### File: `blob.dat`"), "{doc}");
+    assert!(!doc.contains("<Binary file"), "{doc}");
+    assert!(doc.contains("### File: `a.txt`"));
+    assert!(doc.contains("### File: `wide.txt`"), "UTF-16 text stays");
+}
+
+#[test]
+fn header_names_the_directory_not_its_absolute_path() {
+    let dir = project();
+    let out = dir.path().join("out.md");
+    run_with_args(args(dir.path(), &out), Config::default(), &Yes).unwrap();
+    let doc = fs::read_to_string(&out).unwrap();
+    let name = dir
+        .path()
+        .file_name()
+        .unwrap()
+        .to_string_lossy()
+        .to_string();
+    let parent = dir.path().parent().unwrap().to_string_lossy().to_string();
+    assert!(
+        doc.contains(&format!("from the `{name}` directory")),
+        "{doc}"
+    );
+    assert!(!doc.contains(&parent), "absolute path leaked:\n{doc}");
+}
+
+#[test]
+fn binary_prints_summary_with_binary_count_and_clean_header() {
+    let dir = project();
+    let out = Command::new(env!("CARGO_BIN_EXE_context-builder"))
+        .args(["-d", dir.path().to_str().unwrap(), "-o", "-", "-y"])
+        .current_dir(std::env::temp_dir())
+        .output()
+        .unwrap();
+    assert!(out.status.success());
+    let doc = String::from_utf8_lossy(&out.stdout);
+    let err = String::from_utf8_lossy(&out.stderr);
+    assert!(err.contains("1 binary"), "{err}");
+    let parent = dir.path().parent().unwrap().to_string_lossy().to_string();
+    assert!(!doc.contains(&parent), "{doc}");
+    assert!(!doc.contains("<Binary file"));
+}
+
+#[test]
+fn header_keeps_a_symlink_name_and_never_falls_back_to_a_path() {
+    let dir = project();
+    let cwd = dir.path().join("sub");
+    fs::create_dir(&cwd).unwrap();
+    fs::write(cwd.join("x.txt"), "x").unwrap();
+    // `..` resolves to the parent's name; `/` has no name and must not print `/`.
+    let out = dir.path().join("out.md");
+    let mut a = args(&cwd.join(".."), &out);
+    a.input = format!("{}/..", cwd.display());
+    run_with_args(a, Config::default(), &Yes).unwrap();
+    let doc = fs::read_to_string(&out).unwrap();
+    let name = dir
+        .path()
+        .file_name()
+        .unwrap()
+        .to_string_lossy()
+        .to_string();
+    assert!(
+        doc.contains(&format!("from the `{name}` directory")),
+        "{doc}"
+    );
+
+    #[cfg(unix)]
+    {
+        let link = dir.path().join("alias-name");
+        std::os::unix::fs::symlink(&cwd, &link).unwrap();
+        let out2 = dir.path().join("out2.md");
+        run_with_args(args(&link, &out2), Config::default(), &Yes).unwrap();
+        let doc2 = fs::read_to_string(&out2).unwrap();
+        assert!(doc2.contains("from the `alias-name` directory"), "{doc2}");
+    }
+}
+
+#[test]
+fn several_binaries_are_summarized_in_the_plural() {
+    let dir = project();
+    fs::write(dir.path().join("two.dat"), [0u8, 9, 9]).unwrap();
+    let out = Command::new(env!("CARGO_BIN_EXE_context-builder"))
+        .args(["-d", dir.path().to_str().unwrap(), "-o", "-", "-y"])
+        .current_dir(std::env::temp_dir())
+        .output()
+        .unwrap();
+    assert!(String::from_utf8_lossy(&out.stderr).contains("(2 binaries)"));
+}
diff --git a/tests/test_parallel_memory.rs b/tests/test_parallel_memory.rs
index c666b38..f1cc5d0 100644
--- a/tests/test_parallel_memory.rs
+++ b/tests/test_parallel_memory.rs
@@ -210,9 +210,9 @@ fn test_parallel_error_handling() {
     assert!(output_content.contains("fn good1()"));
     assert!(output_content.contains("fn good2()"));
 
-    // Verify binary file is handled with placeholder
-    assert!(output_content.contains("### File: `binary.rs`"));
-    assert!(output_content.contains("<Binary file or unsupported encoding:"));
+    // The binary file is left out of the body and reported under `## Skipped`.
+    assert!(!output_content.contains("### File: `binary.rs`"));
+    assert!(output_content.contains("- `binary.rs` — binary"));
 }
 
 #[cfg(feature = "parallel")]