fix(core): improve JSON get_format_instructions using Opik Agent Optimizer

[Vincent Koc (vincentkoc)]

Description: It's common knowledge that the JSON based get_format_instructions() has a high failure rate on some models. We tested the prompt systematically using JSONSchemaBench with Opik Agent Optimizer (open-source optimizer toolchain) against a few optimizers to improve the underlying prompt using evaluation-based approach.

• The full analysis, test enviroment, and results can be found here.
• We achived a score of 0.97 (+708%) vs. the baseline of 0.12 (orignal prompt) on gpt-4.1.
• This ensures robust schema adherance and can be re-applied to other formats like yaml xml etc.

Results from Opik Optimizer:

Issue: No issues linked
Dependencies: No dependencies

---

• Lint and test: Run make format, make lint and make test from the root of the package(s) you've modified.

[codspeed-hq]

CodSpeed Performance Report

Merging #33718 will not alter performance

_{Comparing vincentkoc:patch-1 (e5cbfc9) with master (a2a9a02)¹}

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

Summary

✅ 13 untouched
⏩ 21 skipped²

Footnotes

1. No successful run was found on master (b5e23e5) during the generation of this report, so a2a9a02 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩

2. 21 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩