Add --topic flag to raw_to_md for filtering off-topic content

When --topic is provided Claude silently discards passages unrelated to
that subject. Without it a generic noise-filter instruction is used.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

raw_to_md.py

@@ -20,11 +20,7 @@ import sys
 
 import anthropic
 
-SYSTEM_PROMPT = """\
+_BASE_GUIDELINES = """\
-You are a technical documentation writer. Your task is to convert raw, unstructured text \
-into a clean, well-structured Markdown document.
-
-Guidelines:
 - Add a ## Summary section at the very top with a 2-3 sentence overview.
 - Organise the rest chronologically when dates, versions, or sequence cues are present; \
   otherwise group by logical topic.
@@ -37,13 +33,34 @@ Guidelines:
 - Output ONLY the Markdown — no preamble, explanation, or trailing commentary.\
 """
 
+_FILTER_GENERIC = (
+    "- Silently discard any passages that are clearly noise, off-topic, or unrelated "
+    "to the main subject of the input (e.g. unrelated chat messages, ads, system logs "
+    "from other contexts). Do not mention the omission."
+)
 
-def convert(raw_text: str, model: str) -> str:
+_FILTER_TOPIC = (
+    "- This document is about: {topic}. "
+    "Silently discard any passages that are clearly unrelated to this topic. "
+    "Do not mention the omission."
+)
+
+
+def _build_system_prompt(topic: str | None) -> str:
+    filter_line = _FILTER_TOPIC.format(topic=topic) if topic else _FILTER_GENERIC
+    return (
+        "You are a technical documentation writer. Your task is to convert raw, "
+        "unstructured text into a clean, well-structured Markdown document.\n\n"
+        f"Guidelines:\n{_BASE_GUIDELINES}\n{filter_line}"
+    )
+
+
+def convert(raw_text: str, model: str, topic: str | None = None) -> str:
     client = anthropic.Anthropic()
     response = client.messages.create(
         model=model,
         max_tokens=8192,
-        system=SYSTEM_PROMPT,
+        system=_build_system_prompt(topic),
         messages=[
             {
                 'role': 'user',
@@ -71,6 +88,11 @@ def main():
         default='claude-sonnet-4-6',
         help='Claude model ID (default: claude-sonnet-4-6)',
     )
+    parser.add_argument(
+        '--topic', '-p',
+        default=None,
+        help='Project topic or description — used to filter out unrelated content',
+    )
     args = parser.parse_args()
 
     if not os.environ.get('ANTHROPIC_API_KEY'):
@@ -95,7 +117,7 @@ def main():
         sys.exit('Error: input is empty.')
 
     print(f'Sending to {args.model}…', file=sys.stderr)
-    result = convert(raw, args.model)
+    result = convert(raw, args.model, topic=args.topic)
 
     if args.output:
         try: