mirror of
https://github.com/4gray/iptvnator.git
synced 2026-10-10 01:56:16 -08:00
4grayandClaude Fable 5
1417632f73
Four P2 findings, each with a regression test: - The authored-body check could never fail. The tag workflow appends GitHub's generated notes to the authored text (FULL_BODY in build-and-make.yaml), so `release.body` is never empty and testing it for emptiness proved nothing. It now compares the body against the local CHANGELOG.md public section — the actual authored source — and reports an internal-only release as such instead of warning about it. - A release with no highlights failed the documented card step, which necessarily included every internal-only release since validation forbids highlights there. Now the hero card is still rendered when public notes exist, an internal-only release writes nothing, and both exit 0. - Card output is keyed by the exact version rather than the minor slug (0.24.0 and 0.24.1 share a blog post but not a card set), and a run first removes the cards a previous run wrote there, so a renamed or dropped highlight cannot leave a stale image waiting to be published. Only files matching what this tool writes are removed. - Telegram could still truncate breaking changes into the counter when a release had no highlights, since the fitting loop trims the tail. Dropping a breaking entry now fails with an actionable error. Docs: docs/architecture/release-pipeline.md, .changes/README.md, CLAUDE.md and AGENTS.md updated for the new output path and the no-highlight and internal-only behavior. 180 tests passing. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>