diff --git a/AGENTS.md b/AGENTS.md
new file mode 100644
index 0000000..ba8a334
--- /dev/null
+++ b/AGENTS.md
@@ -0,0 +1,6 @@
+# Environment
+
+- You are in a nix flake project. When building, add files to git.
+- Run a non-installed tool with `nix run nixpkgs#<package>`.
+- For repeated use, add a tool to development shell.
+- Verify that tools build after with nix build, and run it with nix run.
diff --git a/PLAN.md b/PLAN.md
new file mode 100644
index 0000000..b0c5520
--- /dev/null
+++ b/PLAN.md
@@ -0,0 +1,93 @@
+# org-to-quartz: Org Notes to Quartz Converter
+
+## Project Structure
+
+```
+org-notes-quartz/
+├── flake.nix
+├── pyproject.toml
+├── AGENTS.md
+├── PLAN.md
+└── src/
+    └── org_to_quartz/
+        ├── __init__.py
+        ├── main.py              # CLI (argparse)
+        ├── org_parser.py        # Parse #+key: values + body
+        ├── markdown_writer.py   # YAML front matter + pandoc conversion
+        ├── image_handler.py     # Detect & copy images per-note
+        └── citations/
+            ├── __init__.py
+            ├── resolver.py      # Orchestrator: tries resolvers in order
+            ├── zotero.py        # Better BibTeX JSON-RPC (port 23119)
+            ├── bibtex.py        # pybtex to parse local .bib
+            └── doi.py           # Format as DOI link fallback
+```
+
+## Dependencies
+
+- Python 3.11+
+- pandoc (org → gfm conversion)
+- pybtex (parse .bib files)
+- requests (Zotero JSON-RPC)
+
+## Module Responsibilities
+
+| Module | Responsibility |
+|--------|----------------|
+| `main.py` | CLI: `org-to-quartz <input_dir> <output_dir> [--bib FILE]` |
+| `org_parser.py` | Extract front matter dict + body text from `.org` |
+| `markdown_writer.py` | Build YAML front matter, call pandoc, write `.md` |
+| `image_handler.py` | Find `[[file:img.png]]` / `./img.png`, copy to `<note>/` |
+| `citations/resolver.py` | Chain: Zotero → Bibtex → DOI, return `[Author, Year](zotero://...)` |
+| `citations/zotero.py` | JSON-RPC to `http://localhost:23119/better-bibtex/json-rpc` |
+| `citations/bibtex.py` | Parse `.bib`, format author/year, link to DOI if available |
+| `citations/doi.py` | Format `[cite_key](https://doi.org/...)` or raw key |
+
+## Output Structure
+
+```
+content/
+├── my-note/
+│   ├── index.md          # The converted note
+│   └── diagram.png       # Copied image
+├── another-note/
+│   ├── index.md
+│   └── photo.jpg
+```
+
+## Front Matter Mapping
+
+| Org | YAML |
+|-----|------|
+| `#+title:` | `title:` |
+| `#+filetags:` | `tags:` (list) |
+| `#+date:` | `date:` |
+| `#+hugo_lastmod:` | `lastmod:` |
+| `#+hugo_tags: noexport` | `draft: true` + TODO comment |
+
+## Citation Flow
+
+```
+cite:smith2020  →  Zotero JSON-RPC lookup
+                   ├─ Found → [Smith, 2020](zotero://select/items/@smith2020)
+                   └─ Not found → Bibtex lookup
+                                  ├─ Found → [Smith, 2020](https://doi.org/10.xxx)
+                                  └─ Not found → DOI resolver
+                                                 └─ [smith2020] (raw)
+```
+
+## Implementation Order
+
+1. `flake.nix` + `pyproject.toml` - project skeleton
+2. `org_parser.py` - front matter + body extraction
+3. `markdown_writer.py` - pandoc integration + YAML output
+4. `image_handler.py` - image detection and copying
+5. `citations/bibtex.py` - local bib parsing
+6. `citations/doi.py` - simple fallback formatter
+7. `citations/zotero.py` - JSON-RPC client
+8. `citations/resolver.py` - orchestrator
+9. `main.py` - CLI wiring
+
+## TODO
+
+- [ ] Handle org-roam `id:` links (requires ID→title mapping from all files)
diff --git a/src/org_to_quartz/markdown_writer.py b/src/org_to_quartz/markdown_writer.py
index 22af425..26ae6e4 100644
--- a/src/org_to_quartz/markdown_writer.py
+++ b/src/org_to_quartz/markdown_writer.py
@@ -74,9 +74,9 @@ def process_wikilinks(content: str) -> str:
 def process_citations(content: str, resolver) -> str:
     """Process org-mode citations and resolve them.
 
-    Handles:
+    Handles (after pandoc conversion, brackets may be escaped):
     - cite:key
-    - [cite:@key]
+    - [cite:@key] or \[cite:@key\]
     - [[cite:key][description]]
     """
     if resolver is None:
@@ -86,18 +86,19 @@ def process_citations(content: str, resolver) -> str:
         cite_key = match.group(1)
         return resolver.resolve(cite_key)
 
-    # cite:key pattern
-    content = re.sub(r"cite:([a-zA-Z0-9_-]+)", replace_cite, content)
+    # cite:key pattern (not preceded by @, to avoid double-matching)
+    content = re.sub(r"(?<!@)cite:([a-zA-Z0-9_-]+)", replace_cite, content)
 
-    # [cite:@key] pattern
-    content = re.sub(r"\[cite:@([a-zA-Z0-9_-]+)\]", replace_cite, content)
+    # [cite:@key] pattern (escaped or not after pandoc)
+    # Pandoc escapes [ ] as \[ \]
+    content = re.sub(r"\\?\[cite:@([a-zA-Z0-9_-]+)\\?\]", replace_cite, content)
 
-    # [[cite:key][description]] - use resolved link but keep description context
-    def replace_cite_with_desc(match) -> str:
-        cite_key = match.group(1)
+    # [[cite:key][description]] - after pandoc becomes [description](cite:key)
+    def replace_cite_link(match) -> str:
+        cite_key = match.group(2)
         return resolver.resolve(cite_key)
 
-    content = re.sub(r"\[\[cite:([a-zA-Z0-9_-]+)\]\[[^\]]*\]\]", replace_cite_with_desc, content)
+    content = re.sub(r"\[([^\]]+)\]\(cite:([a-zA-Z0-9_-]+)\)", replace_cite_link, content)
 
     return content