List your company

Environment setup and dependency repair on EnvBench-Python (held-out)

18.6Success RateCODESKILL
4.0927.858511.62515.3915May 25, 2026

Evaluation results

MethodLinks
CODESKILLSkill Backbone=Qwen3.5-4B, Frozen coding policy=Qwen3.5-35B-A3B2026.0518.662.7424.33
SFT-CODESKILLSkill Backbone=Qwen3.5-4B, Frozen coding policy=Qwen3.5-35B-A3B2026.0513.9588.625.25
CODESKILLSkill Backbone=Qwen3.5-4B, Frozen coding policy=GPT-5.4-mini2026.0513.9525.059
Prompt Skill Mgmt.Skill Backbone=GPT-5.4-mini, Frozen coding policy=Qwen3.5-35B-A3B2026.0511.6331.5825
Subtask MemorySkill Backbone=GPT-5.4-mini, Frozen coding policy=Qwen3.5-35B-A3B2026.059.386.3326.5
Subtask MemorySkill Backbone=GPT-5.4-mini, Frozen coding policy=GPT-5.4-mini2026.059.366.7711.81
Prompt Skill Mgmt.Skill Backbone=GPT-5.4-mini, Frozen coding policy=GPT-5.4-mini2026.059.379.610.5
No-skill baselineSkill Backbone=-, Frozen coding policy=Qwen3.5-35B-A3B2026.056.9881.730
Prompt Skill Mgmt.Skill Backbone=Qwen3.5-4B, Frozen coding policy=GPT-5.4-mini2026.056.9831.1910.67
SFT-CODESKILLSkill Backbone=Qwen3.5-4B, Frozen coding policy=GPT-5.4-mini2026.056.9870.7710.5
Prompt Skill Mgmt.Skill Backbone=Qwen3.5-4B, Frozen coding policy=Qwen3.5-35B-A3B2026.054.6591.9526.8
No-skill baselineSkill Backbone=-, Frozen coding policy=GPT-5.4-mini2026.054.6570.7213