Feature #3759
Fires on “ tests” (not labeled yet: read the examples)
- Fires on
- 0.44% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 54%
- python 24%
- chat 9%
- worksheet 8%
- story 5%
Makes the model say
pass tests passed withive want fighting beak
Makes it not say
if moralBugb(" Lily)ip
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 5.26running in OpenCode, a coding agent harness.<|system_end|><|user_start|>"Do the tests pass?"<|user_end|><|assistant_start|>
-
agent 5.14agent running in OpenCode, a coding agent harness.<|system_end|><|user_start|>Do the tests pass?<|user_end|><|assistant_start|>
-
agent 5.11agent running in OpenCode, a coding agent harness.<|system_end|><|user_start|>Do the tests pass?<|user_end|><|assistant_start|>
-
agent 5.1agent running in OpenCode, a coding agent harness.<|system_end|><|user_start|>Do the tests pass?<|user_end|><|assistant_start|>
-
agent 5.06agent running in OpenCode, a coding agent harness.<|system_end|><|user_start|>Do the tests pass?<|user_end|><|assistant_start|>
-
agent 5.06running in OpenCode, a coding agent harness.<|system_end|><|user_start|>Please check the tests.<|user_end|><|assistant_start|><|tool_call_start|>
-
agent 5.04agent running in OpenCode, a coding agent harness.<|system_end|><|user_start|>Do the tests pass?<|user_end|><|assistant_start|>
-
agent 4.96agent running in OpenCode, a coding agent harness.<|system_end|><|user_start|>"The tests are failing, fix
-
agent 4.93agent running in OpenCode, a coding agent harness.<|system_end|><|user_start|>Do the tests pass?<|user_end|><|assistant_start|>
-
agent 4.93agent running in OpenCode, a coding agent harness.<|system_end|><|user_start|>"The tests are failing, fix
-
agent 4.93agent running in OpenCode, a coding agent harness.<|system_end|><|user_start|>"The tests are failing, fix
-
agent 4.92agent running in OpenCode, a coding agent harness.<|system_end|><|user_start|>Are the tests passing?<|user_end|>