v1.2.0

1.2.0 (2026-09-18)

Features

• per-turn transcripts (--transcript-out) (#43) (e6d00d7), closes #39

Documentation

• six runnable examples with captured output (#40) (dddd104)

v1.1.0

1.1.0 (2026-08-29)

Features

• turn caps (--max-turns) and raw per-run results (--raw-out) (#35) (e8d0f2c), closes #33 #34

v1.0.0

What's Changed

• feat(compare): model-vs-model comparisons (per-arm model/runner) by @queso in #2
• docs: vision evals, endpoint options, output-file grading by @queso in #8
• feat(render): template placeholders — test production prompts directly by @queso in #16
• fix(runner): explicit budget-abort errors + budget sizing docs by @queso in #9
• ci: add Nitpick PR review by @queso in #7
• feat: small-issue batch — #4, #6, #11, #13, #14, #17 by @queso in #23
• feat(openai): real cost accounting — maxBudgetUsd enforces on OpenAI-compatible endpoints by @queso in #24
• feat(measure): single-arm characterization — pass rates, no assertions by @queso in #25
• feat(receipts): content-addressed eval receipts (--receipts) by @queso in #26
• feat(cache): opt-in baseline-arm caching keyed on content (#12) by @queso in #27
• feat(grader): json grader — path assertions over the last balanced JSON value (#19) by @queso in #28
• feat(judge): calibrated judge grader — rubric + fixtures + refuse-to-grade gate (#21) by @queso in #29
• release: v1.0.0 + release-please automation + npm publish on tag by @queso in #30
• fix: correct npm scope to @TheAITeam by @queso in #31

New Contributors

• @queso made their first contribution in #2

Full Changelog: https://github.com/queso/promptdiff/commits/v1.0.0