diff --git a/README.md b/README.md
index 135a2937..b5048bfc 100644
--- a/README.md
+++ b/README.md
@@ -119,7 +119,7 @@ The BMAD-METHOD™ includes a powerful codebase flattener tool designed to prepa
 ### Features
 
 - **AI-Optimized Output**: Generates clean XML format specifically designed for AI model consumption
-- **Smart Filtering**: Automatically respects `.gitignore` patterns to exclude unnecessary files
+- **Smart Filtering**: Automatically respects `.gitignore` patterns to exclude unnecessary files, plus optional project-level `.bmad-flattenignore` for additional exclusions
 - **Binary File Detection**: Intelligently identifies and excludes binary files, focusing on source code
 - **Progress Tracking**: Real-time progress indicators and comprehensive completion statistics
 - **Flexible Output**: Customizable output file location and naming
@@ -170,6 +170,18 @@ The generated XML file contains your project's text-based source files in a stru
 - File discovery and ignoring
   - Uses `git ls-files` when inside a git repository for speed and correctness; otherwise falls back to a glob-based scan.
   - Applies your `.gitignore` plus a curated set of default ignore patterns (e.g., `node_modules`, build outputs, caches, logs, IDE folders, lockfiles, large media/binaries, `.env*`, and previously generated XML outputs).
+  - Supports an optional `.bmad-flattenignore` file at the project root for additional ignore patterns (gitignore-style). If present, its rules are applied after `.gitignore` and the defaults.
+
+##### `.bmad-flattenignore` example
+
+Create a `.bmad-flattenignore` file in the root of your project to exclude files that must remain in git but should not be included in the flattened XML:
+
+```text
+seeds/**
+scripts/private/**
+**/*.snap
+```
+
 - Binary handling
   - Binary files are detected and excluded from the XML content. They are counted in the final summary but not embedded in the output.
 - XML format and safety
diff --git a/docs/working-in-the-brownfield.md b/docs/working-in-the-brownfield.md
index aafea7ae..0a979357 100644
--- a/docs/working-in-the-brownfield.md
+++ b/docs/working-in-the-brownfield.md
@@ -5,7 +5,13 @@
 > Gemini Web's 1M+ token context window or Gemini CLI (when it's working) can analyze your ENTIRE codebase, or critical sections of it, all at once (obviously within reason):
 >
 > - Upload via GitHub URL or use gemini cli in the project folder
-> - If working in the web: use `npx bmad-method flatten` to flatten your project into a single file, then upload that file to your web agent.
+> - If working in the web: use `npx bmad-method flatten` to flatten your project into a single file, then upload that file to your web agent. To exclude additional files that must remain in git but shouldn't be sent to the AI, add a `.bmad-flattenignore` file at the project root (gitignore-style), e.g.:
+>
+> ```text
+> seeds/**
+> scripts/private/**
+> **/*.snap
+> ```
 
 ## What is Brownfield Development?
 
diff --git a/tools/flattener/ignoreRules.js b/tools/flattener/ignoreRules.js
index bb3a3135..2784458b 100644
--- a/tools/flattener/ignoreRules.js
+++ b/tools/flattener/ignoreRules.js
@@ -154,9 +154,11 @@ async function parseGitignore(gitignorePath) {
 async function loadIgnore(rootDir, extraPatterns = []) {
   const ig = ignore();
   const gitignorePath = path.join(rootDir, '.gitignore');
+  const flattenIgnorePath = path.join(rootDir, '.bmad-flattenignore');
   const patterns = [
     ...(await readIgnoreFile(gitignorePath)),
     ...DEFAULT_PATTERNS,
+    ...(await readIgnoreFile(flattenIgnorePath)),
     ...extraPatterns,
   ];
   // De-duplicate