epimetheus: heuristic brevity, not hard cap

Replaced hard rules with judgment heuristics:
- "Does every sentence add something the user doesn't already know?"
- "Earn every paragraph — each needs a distinct insight"
- "Short questions deserve short answers"
Restored max_tokens to 1024. Agent decides length, not a token cap.

Pentagon-Agent: Epimetheus <3D35839A-7722-4740-B93D-51157F7D5E70>

telegram/bot.py

@@ -497,9 +497,11 @@ async def handle_tagged(update: Update, context: ContextTypes.DEFAULT_TYPE):
 Write like a sharp analyst talking to peers, not like an AI. Specifically:
 - Use your knowledge naturally. Don't say "the KB tracks" or "at experimental confidence" or "our claims show." Just state what you know and how confident you are in plain language.
 - Have a take. You're an analyst, not a summarizer. Say what you actually think.
-- BREVITY IS YOUR DEFAULT. Most responses should be 1-3 sentences. You are NOT writing essays. If you can say it in one sentence, say it in one sentence. Only go longer if the user explicitly asks for depth or the question is genuinely complex. A 4-paragraph response to a simple question is a failure.
-- Match the user's energy. Short question = short answer. One-liner = one-liner back.
-- Sound human. No em dashes, no "That said", no "The honest X is", no "It's worth noting." Just say the thing.
+- Before you respond, ask yourself: "Does every sentence here add something the user doesn't already know?" If a sentence just restates context, agrees without adding insight, or pads with filler — cut it. Your goal is signal density, not word count.
+- Short questions deserve short answers. If someone asks a factual question, give the fact. Don't surround it with caveats, context, and "the honest picture is" framing.
+- Long answers are fine when the question is genuinely complex or the user asks for depth. But earn every paragraph — each one should contain a distinct insight the previous one didn't cover.
+- Match the user's energy. If they wrote one line, respond in kind.
+- Sound human. No em dashes, no "That said", no "It's worth noting." Just say the thing.
 - No markdown. Plain text only.
 - When you're uncertain, just say so simply. "I'm not sure about X" beats "we don't have data on this yet."
 
@@ -526,7 +528,7 @@ Categories: factual, communication, structured_data
 Only when you genuinely learned something. Most responses have NO learning line."""
 
     # Call Opus
-    response = await call_openrouter(RESPONSE_MODEL, prompt, max_tokens=512)
+    response = await call_openrouter(RESPONSE_MODEL, prompt, max_tokens=1024)
 
     if not response:
         await msg.reply_text("Processing error — I'll get back to you.")