feat: DIA-1511: Support data batch context for autorefinement

[matt-bernstein]

fix /improve-prompt request to not require a jobid, as a job has not yet been created

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 53.33333% with 7 lines in your changes missing coverage. Please review.

Project coverage is 65.70%. Comparing base (b0f1e1d) to head (a6dbc10).

Files with missing lines	Patch %	Lines
adala/skills/collection/prompt_improvement.py	57.14%	3 Missing ⚠️
adala/agents/base.py	0.00%	2 Missing ⚠️
adala/skills/_base.py	60.00%	2 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #232   +/-   ##
=======================================
  Coverage   65.70%   65.70%           
=======================================
  Files          47       47           
  Lines        2350     2350           
=======================================
  Hits         1544     1544           
  Misses        806      806           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[niklub]

Does this PR introduce any functional changes or is it only about reformatting and renaming?

[matt-bernstein]

the only functional change is this

fix /improve-prompt request to not require a jobid, as a job has not yet been created

which I assume was a bug, unless it's to support some future pathway what I didn't know about

[pakelley]

Approved as far as these changes being fine, but to be clear: do we just not actually need any substantive changes (aside from not requiring the job_id) to support this?

[matt-bernstein]

Approved as far as these changes being fine, but to be clear: do we just not actually need any substantive changes (aside from not requiring the job_id) to support this?

That's right. It's a bit hard to follow all the plumbing, but the PromptImprovementSkill is currently implemented as an AnalysisSkill: DataFrame -> Record, where all other inputs go in the static instructions field and the GT data batch is the DataFrame. This needs some rethinking in the future if we want the ability to input multiple kinds of example data, such as with and without GT, or positive (correct) vs negative (incorrect) examples. Especially since we don't actually want anything implemented in the AnalysisSkill (just batching, afaik), But that's what we have for now.