@code_engineering/single_file_debug
cost-benefit threshold at ~300 lines / single-call scope
My Python function returns wrong output on second call
locate@code_engineering/single_file_debug
Sonnet 4.6 sufficient for isolated bugs under ~300 lines. Escalate to Opus 4.8 when bug involves shared state or requires multi-step causal chain.
bug is isolated, context fits one prompt, cost matters
claude/sonnet-4.6bug involves subtle state, multiple suspects, correctness is critical
claude/opus-4.8you need precise root-cause location, not symptom description
openai/gpt-5.5terminal-heavy debugging workflow
openai/gpt-5.5| ID | Relation | Actor | Result | Output |
|---|---|---|---|---|
| CT-D1-001 | identify@single_file_debug | claude/opus-4.8 | accepted | classifies bug type accurately, including subtle state issues |
| CT-D1-002 | identify@single_file_debug | claude/sonnet-4.6 | accepted | matches Opus for common patterns; preferred by devs |
| CT-D1-003 | identify@single_file_debug | openai/gpt-5.5 | bounded | accurate classification but sometimes stops at symptom level |
| CT-D1-004 | validate@single_file_debug | claude/opus-4.8 | accepted | more likely to flag own code flaws vs prior Opus; stronger dishonest-summary resistance |
| CT-D1-005 | validate@single_file_debug | claude/sonnet-4.6 | bounded | reliable for isolated bugs; can miss subtle flaws silently vs Opus 4.8 |
Opus 4.8 is 4x less likely to miss code flaws than Opus 4.7
E-GLOBAL-004Sonnet 4.6 preferred by developers over prior flagship Opus 4.5 in controlled preference testing
Anthropic Sonnet 4.6 release, 2026GPT-5.5 weaker than Opus 4.8 on precise root-cause location vs symptom description in debugging
inferred from SWE-Bench Pro gap