Skip to content

Diffly CLI v1.0.0

Choose a tag to compare

@VIVAAN-DHAWAN VIVAAN-DHAWAN released this 22 Aug 17:26
· 17 commits to main since this release

Diffly CLI v1.0.0

Diffly CLI 1.0.0 is the first production-ready release of deterministic pull-request triage. It turns pull-request evidence—changed files, symbols, direct callers, tests, checks, and risk signals—into a clear review decision without making the verdict depend on an LLM.

Highlights

  • Verdicts that match the evidence. PASS is now the normal result for healthy pull requests. QUARANTINE is reserved for concrete review gates such as security-sensitive code, database schema or migration changes, dependency changes, or pending checks. BLOCK is reserved for failed required checks and high-confidence credential exposure.
  • Reliable explanations. When you opt in, Diffly always includes an explanation in the review: a bounded, redacted AI narrative when a provider is configured, or a clearly labelled local explanation when it is not. Explanations never alter the deterministic verdict.
  • A focused terminal workflow. The guided wizard, loading transition, keyboard-driven review, and report output have been redesigned for a centred, readable terminal experience with clearer prompts and recovery messages.
  • Works where you work. Review GitHub pull requests, analyze local changes offline, consume stable JSON in CI, or use the bundled GitHub Action. Diffly also reuses your authenticated GitHub CLI session when available.

Upgrade

pip install --upgrade diffly-cli

Or run diffly update from an existing 0.4.0+ installation.

Confidence

Version 1.0.0 is backed by the project test suite and cross-version CI coverage for Python 3.10–3.13. Diffly is production-ready as a deterministic review signal: it helps teams focus their review effort, but it does not replace human review or guarantee that a pull request is defect-free.

Thank you to everyone who tested Diffly before 1.0.0. This release establishes a dependable foundation for reviewing pull requests with more context and less noise.