[jan-auer]

Agents often wrote changelog entries that described the implementation instead of what changes for users. They also missed PRs that should be excluded. The new changelog-entry skill guides them through this. The Craft workspace details move from AGENTS.md into the skill. AGENTS.md now points to the skill, and says to mark breaking changes with ! in the PR title.

[github-actions]

Semver Impact of This PR

⚪ None (no workspace changed)

📋 Changelog Preview

This is how your changes will appear in the changelog.
Entries from this PR are highlighted with a left border (blockquote style).

---

This PR does not change any workspace, so it will not appear in any changelog.

---

_{🤖 This preview updates automatically when you update the PR.}

[jan-auer]

Here's an overview of what the new skill will do to the currently open PRs:

Excluded with #skip-changelog: these change internals, or code that users can't reach or configure yet.

PR	Reason
#747 generic garbage collector	Internal building block
#748 garbage collector configuration	Not wired up until MultiChangeStream lands; the PR that enables the GC gets the entry
#653 SQLite-backed GcStorage	Internal, not configurable
#643 MultiStream ChangeStream	Not used yet
#689 RocksDB backend	Not selectable in the server config
#426 generated options schema	Proof of concept, internal

New or changed ### Changelog Entry: the title doesn't describe the change well for users.

• fix(local-fs): Remove empty directories after deleting objects #755: the title doesn't say which backend.

  - Remove empty directories when deleting objects from the filesystem backend

• feat(service): Add CQL high-volume backend for Cassandra and ScyllaDB #675 / feat(storage): Add CQL high-volume backend #676: alternative CQL backends, so the same entry works for whichever lands.

  - Add a high-volume backend for Cassandra and ScyllaDB, configured with `type: cql`
    - It's experimental and has only been tested against single-node clusters.

• feat(service): Implement BigTable-backed ChangeLog for tiered storage crash recovery #413: the title describes the mechanism, not what users get.

  - Clean up orphaned long-term blobs after a crash during a tiered storage write
    - Turn this on with the new `changelog` option on tiered storage. It defaults to `noop`.

• feat(service): Add changelog heartbeats and continuous recovery loop #414:

  - Recover from crashed tiered storage writes continuously instead of only at startup

• feat(py-client): Implement "many" api for batch requests #546: the entry is good, but it should move to the end of the description.

Breaking change: #750 changes the on-disk layout of the filesystem backend with no migration. If that counts as breaking, the title becomes fix(local-fs)!: and the entry gets a sub-bullet:

- Encode object keys in the filesystem backend so they stay inside the storage root
  - Objects stored by earlier versions can no longer be read. There is no migration.

No change: #550 and this PR only change files outside the release workspaces, so no changelog lists them.

What this shows: most of the value is in excluding PRs. Without the marker, six internal PRs would show up as noise in the server changelog. Where the skill rewrites an entry, it mostly adds the user-facing context the title leaves out: which backend, how to turn it on, and what users have to do.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.82%. Comparing base (c678417) to head (753ffa6).
⚠️ Report is 4 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #761      +/-   ##
==========================================
- Coverage   91.91%   91.82%   -0.10%     
==========================================
  Files         119      119              
  Lines       22771    22837      +66     
==========================================
+ Hits        20930    20969      +39     
- Misses       1841     1868      +27     

Components	Coverage Δ
Rust Backend	94.80% <ø> (ø)
Rust Client	89.13% <ø> (ø)
Python Client	91.64% <ø> (-2.26%)	⬇️

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.