since-cutoff 0.4.0: scan leads with your code, notes from the API diff, sync and status

[MohammadHijjawi97]

since-cutoff 0.4.0 is on PyPI (uvx since-cutoff@latest scan). It changes what the free scan gives you, before any model is involved.

scan leads with your code

The first line is now "Your code uses N APIs that changed after 's training cutoff", followed by each API: what changed, where your code uses it, and a ready note for your agent. Everything else is one line, and --all expands it. No model calls, no API key.

Notes from the API diff, each saying what was checked

Every note carries a tag:

• [diff]: the change is in the static comparison of the two releases.
• [diff + library]: the replacement is named in the library's own deprecation text and exists in your pinned version.
• [diff + move checked]: the object at the new path keeps most of the old one's members.
• [diff; probable rename]: same position and annotation, new name, labelled as a guess.
• [type-checked]: written during run, kept because its example type-checks against your version.

Names that merely look similar are never written as replacements; --suggestions adds them, tagged [not confirmed]. Library advice is quoted as advice, not presented as a replacement. "verified" is no longer used on its own anywhere.

sync and status keep the notes current

since-cutoff sync writes and maintains the notes block from your lockfile: it adds notes for newly used changed APIs, updates bumped packages, drops what no longer applies (and says why), shows a unified diff, and asks before writing. sync --check for CI and pre-commit; since-cutoff status tells you offline whether the block matches the lockfile. The Claude Code plugin gets a session-start hook that says so too.

Also

Matching now finds re-exported names (from huggingface_hub import hf_hub_download) and every path of a change; a beta-only change is no longer attributed to the non-beta API. The GitHub Action gained check-notes, --fail-on old-form and GitHub annotations. 990 tests, 94% branch coverage.

Line numbers for each use are tracked in #8; writing to both AGENTS.md and CLAUDE.md is #13 (a good first issue). Full changelog: https://github.com/MohammadHijjawi97/since-cutoff/blob/main/CHANGELOG.md

Next: the head-to-head benchmark. The same coding tasks, with real tests, run by Claude Code alone, with Context7 docs, with since-cutoff's notes, and with both. Results will be published whatever they show.