rbumg/evalgate

Eval regression gate: run golden-dataset evals against an OpenAI-compatible endpoint, compare to a committed baseline, and fail the job on regression. Writes a markdown report to the job step summary.

View on GitHub

Trust Signals

Scorecard Score
not yet scored
Maintenance Recency
Stale
License
None
namedescriptionrequireddefault
suitesSuite file or directory, relative to the workspace.noevals
baselinePath to the committed baseline file.nobaseline.json
endpointBase URL of the OpenAI-compatible endpoint.nohttps://api.openai.com/v1
modelModel to evaluate.yes
api-keyAPI key for the endpoint. Pass a repository secret.yes
judge-endpointBase URL for the judge model. Defaults to endpoint.no""
judge-modelJudge model for llm_judge checks. Defaults to model.no""
judge-api-keyAPI key for the judge endpoint. Defaults to api-key.no""
toleranceAllowed aggregate score drop before failing.no0.01
repeatRuns per case, aggregated to damp flakiness.no1
extra-argsExtra flags passed verbatim to "evalgate run".no""

no outputs