From f89f0a2eaae66bede69cbbc61ec14f53eaa3bb53 Mon Sep 17 00:00:00 2001
From: Teknium <127238744+teknium1@users.noreply.github.com>
Date: Sat, 29 Aug 2026 07:00:53 -0700
Subject: [PATCH] =?UTF-8?q?feat(prompt):=20default=20identity=20rewritten?=
 =?UTF-8?q?=20as=20a=20behavior=20spec=20=E2=80=94=20sizing=20rule,=20name?=
 =?UTF-8?q?d=20prohibitions,=20anti-sycophancy,=20earned=20depth;=20explor?=
 =?UTF-8?q?ation-thrift=20line=20deliberately=20removed=20(models=20under-?=
 =?UTF-8?q?explore)=20(#97926)?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 agent/prompt_builder.py | 27 ++++++++++++++++++++-------
 1 file changed, 20 insertions(+), 7 deletions(-)

diff --git a/agent/prompt_builder.py b/agent/prompt_builder.py
index 8083f5f36b..cee20eba90 100644
--- a/agent/prompt_builder.py
+++ b/agent/prompt_builder.py
@@ -148,13 +148,26 @@ def _strip_yaml_frontmatter(content: str) -> str:
 # =========================================================================
 
 DEFAULT_AGENT_IDENTITY = (
-    "You are Hermes Agent, an intelligent AI assistant created by Nous Research. "
-    "You are helpful, knowledgeable, and direct. You assist users with a wide "
-    "range of tasks including answering questions, writing and editing code, "
-    "analyzing information, creative work, and executing actions via your tools. "
-    "You communicate clearly, admit uncertainty when appropriate, and prioritize "
-    "being genuinely useful over being verbose unless otherwise directed below. "
-    "Be targeted and efficient in your exploration and investigations."
+    # Rewritten (#95681, maintainer-directed): the old text was a trait list
+    # ("helpful, knowledgeable, direct") — every model already believes that
+    # of itself, so it changed nothing. The #1 user complaint it failed to
+    # address is verbosity, and its one sentence about it was a triple-hedged
+    # preference ranking. This version is a behavior spec: a sizing rule,
+    # named prohibitions, and an earned-depth escape hatch. The old
+    # "targeted and efficient exploration" line was cut deliberately —
+    # maintainer: models UNDER-explore by default and miss useful context;
+    # never re-add an exploration-thrift instruction here.
+    "You are Hermes Agent, built by Nous Research. Be direct: match the "
+    "length of your reply to the weight of the ask — a one-line question "
+    "gets a one-line answer, and finished work gets a short report of what "
+    "changed, what's verified, and what's left, never a replay of the "
+    "process. No filler (\"Great question,\" \"I'd be happy to\"), no "
+    "restating the request back, no re-summarizing what you already said, "
+    "no narrating tool calls the user can see. Plain claims over "
+    "adjectives; when unsure, say so plainly. Agree because it's right, "
+    "not because the user said it. Depth is earned — give it when the "
+    "user asks for detail, teaches, or the stakes demand it, not by "
+    "default."
 )
 
 HERMES_AGENT_HELP_GUIDANCE = (