fix(release): tidy changelog + per-nightly changelog in hub release

[insipx]

Two changelog improvements, both stemming from the scruffy first generated Release PR (#3712).

1. Tidy the generated changelog (cliff.toml)

Body-dump scruff — the template rendered commit.message (full message = subject + body). GitHub squash-merges fill the commit body with the PR description, so each merge dumped its whole PR body into the changelog: Macroscope HTML-comment summary blocks, Co-authored-by: trailers, even a pasted Nix build-error log (~316 of #3712's 461 lines).

-{{ commit.message | upper_first }}
+{{ commit.message | split(pat="\n") | first | trim | upper_first }}

Fake per-scope groups — non-conventional commits (e.g. xmtp_proto: …, bindings/mobile: …) made git-cliff invent junk section headers like Xmtp_proto / Bindings/mobile. A catch-all collects them into one Other group:

+  { message = ".*", group = "<!-- 8 -->Other" },

Result (regenerated from main): group headers are exactly Features · Bug Fixes · Performance · Documentation · Other, every entry a clean subject line, zero scruff.

2. Per-nightly changelog in the hub release (release.yml)

Each nightly hub GitHub Release now appends a "Changes since the last nightly" section — the git-cliff delta over <previous-nightly-tag>..HEAD (--strip header), computed before this run is tagged so the newest existing v*-nightly.* tag is genuinely the prior one. First-ever nightly (no prior nightly tag) falls back to since-last-stable. Empty output is dropped. Adds git-cliff to the hub job.

Uses an explicit <tag>..HEAD range rather than the config's stable-anchored tag_pattern, so it diffs against the last nightly, not the last stable. Relies on the §1 fixes for clean output.

Verification

• Regenerated changelog from main: clean, 5 well-formed groups, no scruff.
• Simulated the per-nightly delta (<tag>..HEAD --strip header): clean grouped output, 22 bullets, no header.
• All workflow YAML validated.

Once merged, the open Release PR (#3712) regenerates clean on the next push to main, and the next real nightly carries a proper since-last-nightly changelog.

🤖 Generated with Claude Code

Note

Tidy generated changelog to show subject-only commits and group non-conventional commits into 'Other'

• Updates cliff.toml to trim commit messages to their first line only and capitalise the first character, removing multi-line noise from changelogs.
• Adds a catch-all parser rule in cliff.toml that places non-conventional commits into an 'Other' group instead of creating per-scope groups.
• Updates release.yml to install git-cliff, compute a delta changelog from the previous nightly tag to HEAD, and append a

^{Macroscope summarized eecd171.}

[claude]

Claude finished @insipx's task —— View job

---

Code Review — 2026-06-03 22:14 UTC

Todo List:

• Read changed files
• Analyze cliff.toml changes
• Analyze workflow changes
• Provide review feedback

---

Summary

The changes effectively solve both stated problems with minimal, well-targeted fixes. Code quality is high.

Detailed Review

cliff.toml (lines 22, 63)

✅ Subject-line extraction (line 22):

{{ commit.message | split(pat="\n") | first | trim | upper_first }}

Clean solution. The pipeline correctly isolates the subject line from squash-merge bodies.

✅ Catch-all rule (line 63):

{ message = ".*", group = "Other" },

Proper placement at the end of commit_parsers. The `` sort prefix ensures "Other" appears last, matching the documented PR behavior.

Minor observation: No issues, but consider whether "Miscellaneous" (line 57, ) and "Other" (line 63, ) could cause user confusion about which bucket applies. The distinction is clear in the rules (conventional chore/ci/build → Miscellaneous; non-conventional → Other), but naming overlap might be subtle for humans reading changelogs. Not blocking—works correctly as-is.

.github/workflows/release.yml (lines 421-437)

✅ Per-nightly changelog logic:
The new section computes the delta since the previous nightly tag, which is a solid UX improvement for nightly releases. Implementation details:

• Line 426: git tag -l 'v*-nightly.*' --sort=-creatordate | head -1 correctly finds the most recent prior nightly
• Lines 429, 432: Appropriate fallback to --unreleased for first-ever nightlies
• Lines 435-437: Empty-content guard prevents publishing whitespace-only sections

🔍 Edge case: Line 429 uses 2>/dev/null || true to suppress git-cliff errors. If git-cliff fails (bad config, missing dependencies), the changelog silently becomes empty. Consider whether a failed git-cliff invocation should:

1. Continue silently (current behavior, fail-open)
2. Fail the workflow (fail-closed to catch config breakage)

Current approach is reasonable for robustness, but if git-cliff breaks due to the cliff.toml changes, you won't know until you notice the empty changelog in a release. Trade-off between resilience and observability.

Lines 469-475: The conditional append of nightly changelog to the release body is well-structured. The heredoc formatting and test [ -s /tmp/nightly-changelog.md ] are correct.

---

Test Coverage

No test changes included. Changelog generation is typically verified manually (as done in the PR description). For a config-only change, this is acceptable. Consider whether any of the following would add value:

• A CI job that runs git cliff on a test repo/commits to validate the config parses
• Smoke test that the split(pat="\n") | first template renders without errors

Given the manual verification shown in the PR body ("regenerated from main" with before/after), this is not a blocker.

---

Performance & Security

• Performance: No concerns. Template filters and regex matching add negligible overhead to changelog generation.
• Security: No concerns. Changes are to build/release tooling configuration, not runtime code.

---

Conclusion

Changes are well-designed and address the stated problems. No blocking issues. The one observability consideration (git-cliff silent failure) is a design choice, not a bug.

---

[macroscopeapp]

Approvability

Verdict: Approved

Minor CI/CD configuration changes to improve changelog generation for nightly releases. The changes only affect release note formatting, not production behavior, and the author owns both modified files.

^{You can customize Macroscope's approvability policy. Learn more.}

Dismissing prior approval to re-evaluate dc5af90

Dismissing prior approval to re-evaluate eecd171