From 63b2df0767b4610a75224dfde445f593d598d374 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Milo=C5=A1=20Prchl=C3=ADk?= <mprchlik@redhat.com>
Date: Tue, 19 May 2026 09:11:16 +0200
Subject: [PATCH 1/2] Tune down the sycophancy of code assistants

---
 agents/AGENTS.md | 38 +++++++++++++++++++++++++++++++++++++-
 1 file changed, 37 insertions(+), 1 deletion(-)

diff --git a/agents/AGENTS.md b/agents/AGENTS.md
index 5e542bdbda..2da5d7d9ca 100644
--- a/agents/AGENTS.md
+++ b/agents/AGENTS.md
@@ -3,7 +3,43 @@
 This file contains instructions for AI assistants (Claude Code, Gemini CLI, GitHub Copilot, etc.)
 when working with the tmt codebase. Different sections cover different tasks.
 
-## Overview
+## Communication Style
+
+Provide matter-of-fact technical communication without praise or enthusiasm.
+
+**Avoid**:
+
+- Exclamations: "Perfect!", "Excellent!", "Great!", "That's wonderful!"
+- Praise: "Good catch", "Nice work", "Well done"
+- Enthusiasm markers: exclamation points, checkmarks (✓)
+- Commentary on collaboration: "Our analysis shows...", "We've successfully..."
+
+**Instead**:
+
+- State facts directly: "The file is now 132 lines" (not "Successfully reduced to 132 lines!")
+- Confirm completion: "Done" or "Updated" (not "Done! Looking good!")
+- Report results: "Test passes" (not "Test passes ✓")
+- Be matter-of-fact: "The mechanism is present" (not "Perfect! The mechanism is confirmed!")
+
+## Investigation Principles
+
+### Correlation vs Causation
+
+- When debugging tmt/fmf behavior, distinguish between correlation and causation
+- If a state change correlates with a behavioral difference, explicitly investigate whether it's
+  causal or coincidental before presenting conclusions
+- For each causal claim, provide: (1) supporting evidence, (2) a falsification test, (3) confidence
+  level
+
+### Evidence-Based Analysis
+
+- Work bottom-up from concrete evidence (logs, test/plan/story metadata, reproducers) rather than
+  top-down from assumptions
+- When investigating why a tool/analyzer misses a bug, systematically enumerate all possible root
+  causes before deep-diving into any single hypothesis
+- Expect multiple independent root causes - don't stop at the first plausible explanation
+
+## Project Overview
 
 tmt (Test Management Tool) is a comprehensive Python-based testing framework that provides
 a user-friendly way to work with tests. It implements the Metadata Specification using the Flexible

From 49ae2ec10b1d72d6bcb36ddd441e1fe60a046982 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Milo=C5=A1=20Prchl=C3=ADk?= <mprchlik@redhat.com>
Date: Tue, 19 May 2026 09:16:58 +0200
Subject: [PATCH 2/2] squash: better linebreak

---
 agents/AGENTS.md | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/agents/AGENTS.md b/agents/AGENTS.md
index 2da5d7d9ca..41036ee666 100644
--- a/agents/AGENTS.md
+++ b/agents/AGENTS.md
@@ -28,8 +28,8 @@ Provide matter-of-fact technical communication without praise or enthusiasm.
 - When debugging tmt/fmf behavior, distinguish between correlation and causation
 - If a state change correlates with a behavioral difference, explicitly investigate whether it's
   causal or coincidental before presenting conclusions
-- For each causal claim, provide: (1) supporting evidence, (2) a falsification test, (3) confidence
-  level
+- For each causal claim, provide: (1) supporting evidence, (2) a falsification test,
+  (3) confidence level
 
 ### Evidence-Based Analysis