hidai25/eval-view

Minor release — 33 commits since 0.6.2, 14 new user-facing features.

Highlights

• Aider CLI adapter — drive Aider as an EvalView adapter
• Autopr loop — prod-incident → regression-test → PR, closed loop
• Flake quarantine — known-flaky tests don't block CI, with governance metadata
• Release verdict + evalview since — graded ship/hold verdict + change brief
• progress / drift / slack-digest — investigative loop commands
• Noise confirmation gate + --strict bypass — two-cycle rule before alerting
• Slow-agent warning — real wall-clock latency regression detection
• Observability signals — trust score, tool-loop, brittle-recovery, gaming checks
• Improvement recommendation engine — prioritized stabilize / tighten / add-check suggestions
• Simulation harness + decision-rationale (schema v2) — scripted multi-turn scenarios, machine-readable reasons
• snapshot --json — CI-friendly, hardened for edge cases
• check --explain — deep trace narrative for root-cause hypotheses
• Token cost breakdown in check — input/output/cached tokens + cost delta vs baseline
• Skill-doctor char-budget refinement — disable-model-invocation skills excluded

Plus ~10 fixes (mypy narrowing, dogfood hardening, slack-digest type errors, noise strict-bucket leak, snapshot --json CI hardening) and README/CLI doc improvements.

Install

pip install evalview==0.7.0
# or
npm install [email protected]

Full changelog: https://github.com/hidai25/eval-view/blob/v0.7.0/CHANGELOG.md