diff0 v0.1.4

diff0 v0.1.4 makes comparison gates more reliable and includes the library API preview.

• Performance budgets use unrounded measurements, correctly detecting increases just above a configured threshold.
• Inconclusive behavioral observations remain in report details without independently triggering yellow verdicts or drift enforcement.
• Node.js 20 support is restored through Commander 14. The published CLI excludes the repository's Eve maintenance runtime.
• Library entrypoints cover comparison, execution, Eve adaptation, and report rendering.
• Getting Started documents baseline, eval, and credential setup. The website preserves the original September 4 showcase evidence separately from the latest live comparison.

Install or run:

npx @knowbody/diff0@0.1.4 run --base main

GitHub Actions:

- uses: knowbody/diff0/action@v0.1.4

The CLI supports Linux/macOS and Node.js 20+. Current Eve applications require Node.js 24+. The end-to-end suite targets Eve 0.52.2.

diff0 v0.1.3

What changed

• Add directional median performance budgets for cost, input tokens, output tokens, and duration.
• Add granular typed enforcement categories for eval, score, performance, behavioral, and comparison-validity findings.
• Report the actual Eve sandbox as unknown and separate the host-default candidate.
• Detect authored sandbox changes and user-configured validity paths.
• Fix estimate execution controls, CLI usage exits, the deterministic demo, Action enforcement, and report claims.
• Align the CLI, GitHub Action, documentation, examples, and website marketing.

Compatibility

Legacy fail-on policies remain supported. JSON report schema is now 4 and cache schema is now 6.

diff0 v0.1.2

A hardening release focused on making the public comparison guarantees match actual behavior.

What changed

• Reject dirty local HEAD comparisons instead of silently testing uncommitted state.
• Flag eval-harness changes and comparison-validity mismatches.
• Detect checkout v7 persisted credentials and recover from stale PR comments.
• Counterbalance AB/BA execution order to reduce order bias.
• Clarify limitations, the default yellow verdict, and spend controls.
• Improve website navigation, accessibility, metadata, and SEO.

Try it

npx @knowbody/diff0@0.1.2 run --base main

For GitHub Actions, pin knowbody/diff0/action@v0.1.2.

Validation

• 321 unit tests and 6 integration tests
• package install and CLI smoke test
• minimum-supported Node 20 smoke test
• Action bundle verification
• production website build
• 10-run-per-ref real-model dogfood comparison with no confirmed regression or comparison-validity mismatch

diff0 v0.1.1

A cleaner public release of diff0.

What changed

• Newcomer-first README and marketing site with a one-command quickstart
• Real-model PR showcase backed by 10 runs per revision
• Fixed the npm CLI executable and added diff0 --version
• Reduced the bundled GitHub Action from 610 KB to 275 KB
• Updated examples to current GitHub Actions releases
• Clarified credentials, security, run filters, and trust boundaries

Try it

npx @knowbody/diff0 run --base main

Full validation includes 314 unit tests, 6 integration tests, package-install smoke tests, the production website build, and a live Claude comparison.

diff0 v0.1.0

Initial public release of diff0.

• Compare an Eve agent across two Git revisions.
• Detect eval, tool, skill, subagent, output, cost, and latency drift.
• Run locally or as a bundled GitHub Action with sticky PR reports.
• Exercise the complete pipeline with deterministic fixtures and integration tests.

Install with: pnpm add -D @knowbody/diff0