docs: backfill and update changelog #10

Merged
rcsheets merged 4 commits from docs/changelog-backfill into main 2026-07-21 01:41:14 +00:00
Owner
No description provided.
"Release entries below the seed" pointed the wrong way: the seed is 0.0.0 at
the bottom of the file and generated entries are spliced in above it. The
phrasing was ambiguous besides, since "below" could mean either later in
version order or further down the page, and here those point in opposite
directions.

Name the exception rather than a direction, and state the ordering outright.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The entries reading "STUB: N bytes of changes." date from before SLP was
deployed, when the script could not call the inference service it needed
because the service is one SLP itself provides. hack/backfill-changelog.sh
regenerates them from the same commit ranges.

The summarizer moves to hack/changelog-lib.sh, sourced by both scripts,
specifically so the two cannot drift: an entry written at release time and one
filled in years later should read as though the same hand wrote them, which
stops being true the moment there are two copies of the prompt.

Backfill only rewrites sections whose entire body is a STUB line, so it never
overwrites real prose and is safe to re-run. There is no --dry-run flag: the
file is always in git, so `git diff` reviews the actual spliced result and
`git checkout` discards it, which is strictly better than printing loose
summaries that skip the splice.

Attribution trailers are now filtered alongside Co-Authored-By, because the
early history is dense with "Generated by <tool>." sentences. Who or what
wrote a commit has no place in an entry describing what changed, and in bulk
it is enough of the input to skew a summary toward tooling.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The 180s per-attempt ceiling was tight enough that summarizing the whole
pre-1.0 history in one request landed on either side of it depending on load:
it succeeded on a retry once and then failed all three attempts. Time to first
token scales with how much history is in the request, and the first release is
by far the largest range there will ever be.

Default 600s, overridable with INFERENCE_TIMEOUT. Release ranges are small and
answer in seconds, so this bound is really for the backfill.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
docs(changelog): backfill the entries left as stubs
All checks were successful
ci / check (pull_request) Successful in 46s
a52e0f02bd
Replaces the "STUB: N bytes of changes." placeholders for 0.1.0 through 0.4.0
with real summaries, generated by hack/backfill-changelog.sh from the same
commit ranges and the same summarizer the release pipeline uses.

Drops the header note about backfill being pending, which is no longer true.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
rcsheets deleted branch docs/changelog-backfill 2026-07-21 01:41:14 +00:00
Sign in to join this conversation.
No reviewers
No labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
brooktrails/slp!10
No description provided.