Feature #2910

Strongest examples

On text the dictionary never trained on. The underlined token is where it fires most.

1. agent 2.44
   coding agent harness.<|system_end|><|user_start|>"Can you run demo.py?"<|user_end|><|assistant_start|><|tool_call_start|>she
2. agent 2.35
   greeting.py that prints "Welcome!", then run it.<|user_end|><|assistant_start|><|tool_call_start|>write
3. agent 2.31
   then run main.py.<|user_end|><|assistant_start|><|tool_call_start|>gre
4. agent 2.3
   a coding agent harness.<|system_end|><|user_start|>Can you run main.py?<|user_end|><|assistant_start|><|tool_call_start|>she
5. agent 2.26
   ] for VALUES in demo.py, then run demo.py.<|user_end|><|assistant_start|><|tool_call_start|>read<|arg|>
6. agent 2.26
   Change LIMIT to 12 and run main.py.<|user_end|><|assistant_start|><|tool_call_start|>gre
7. agent 2.25
   coding agent harness.<|system_end|><|user_start|>"Can you run main.py?"<|user_end|><|assistant_start|><|tool_call_start|>she
8. agent 2.24
   <|system_end|><|user_start|>make say.py print "Hello, Kai!".<|user_end|><|assistant_start|><|tool_call_start|>write
9. agent 2.24
   coding agent harness.<|system_end|><|user_start|>"Can you run demo.py?"<|user_end|><|assistant_start|><|tool_call_start|>she
10. agent 2.22
    "python" in main.py, then run main.py.<|user_end|><|assistant_start|><|tool_call_start|>read<|arg|>
11. agent 2.21
    Set LIMIT to 9, then run main.py please"<|user_end|><|assistant_start|><|tool_call_start|>gre
12. agent 2.21
    <|user_start|>"main.py crashes, can you fix it?"<|user_end|><|assistant_start|><|tool_call_start|>she