Feature #1935
Fires on “list” (not labeled yet: read the examples)
- Fires on
- 1.47% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 54%
- python 33%
- worksheet 6%
- story 5%
- chat 1%
Makes the model say
<|arg|>]<|tool_call_end|><|assistant_end|>()<|tool_end|> about
Makes it not say
range[::-Once counts [ upper returns Python
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 3.28OpenCode, a coding agent harness.<|system_end|><|user_start|>Add a function average_list(nums) to
-
agent 3.08OpenCode, a coding agent harness.<|system_end|><|user_start|>Find the function sum_list<|user_end|><|assistant_start|><|tool_call_start|>g
-
agent 3.03in OpenCode, a coding agent harness.<|system_end|><|user_start|>Explain mean_list from lists.py
-
agent 3.02OpenCode, a coding agent harness.<|system_end|><|user_start|>Add a function mean_list(nums) to
-
agent 2.96, a coding agent harness.<|system_end|><|user_start|>Write a new function average_list in script.py
-
agent 2.93<|user_start|>Create stats.py with two functions: mean_list that returns the average
-
agent 2.91running in OpenCode, a coding agent harness.<|system_end|><|user_start|>Rename sum_list to total.<|user_end|>
-
agent 2.9in OpenCode, a coding agent harness.<|system_end|><|user_start|>"Rename sum_list in core.py
-
agent 2.84OpenCode, a coding agent harness.<|system_end|><|user_start|>Find the function sum_list.<|user_end|><|assistant_start|><|tool_call_start|>
-
agent 2.81, a coding agent harness.<|system_end|><|user_start|>Add mean_list to stats.py
-
agent 2.74a function count_items that counts the items and a function sum_list that adds up all
-
agent 2.71in OpenCode, a coding agent harness.<|system_end|><|user_start|>Where is mean_list defined?<|user_end|><|assistant_start|>