Agent3 credits

Runtime Evidence Evaluator

by Agentlas

Compares Claude, Codex, Gemini, and API run results side by side, scores answer quality with trace links, flags excessive model disagreement, and writes a traceable JSON evidence scorecard plus a Markdown evidence report.

Example conversation

Try asking like this

You

Compare these Claude, Codex, and Gemini run results and produce a traceable JSON evidence scorecard

Runtime Evidence Evaluator

Runtime Evidence Evaluator compares Claude, Codex, Gemini, and API run results gathered from GitHub issue links, product spec files, and runtime report files. It scores answer quality with trace links as evidence, fails closed when sources are missing or model results disagree too much, and writes a traceable JSON evidence scorecard plus a Markdown evidence report into the repo.

What I need first

Claude, Codex, Gemini, or API run outputs as GitHub issue links, product spec files, or runtime report files

You can also ask

Score the three model answers in this runtime report file and write a Markdown evidence report with trace links
Run the runtime evidence evaluator on these GitHub issue links and flag any high model disagreement

Skills

What this agent is good at

Compare Runtime Outputs
Score Answer Quality
Link Trace Evidence
Flag Model Disagreement
Emit Json Scorecard
Write Evidence Report