Released changesets ledger

This directory holds a per-branch record of which changesets have already been released. Each file is named after a release branch (with / replaced by -, e.g. main.txt, release-10.0.txt) and contains one consumed changeset id per line.

Why this exists

Fixes are cherry-picked between main and release/* branches, so the same changeset file can end up on both. The release branch's release consumes its copy and deletes it, but the cherry-picked copy on main would otherwise survive the merge back and be applied a second time on the next main release.

The ledger prevents that double-application. Each branch only writes to its own file, so cross-branch merges are conflict-free; the wrapper around changeset version reads the union of every file when deciding what to skip.

The directory lives at the repo root (sibling of .changeset/) rather than inside .changeset/ because @changesets/read treats every directory inside .changeset/ as a legacy v1 changeset and tries to read changes.md from it.

How it gets updated

pnpm bump runs the wrapper at __utils__/scripts/src/bump.ts, which:

1. Reads every *.txt in this directory and unions the ids.
2. Hides any .changeset/<id>.md whose id is in that union (renames to <id>.md.released) so changeset version does not see it.
3. Runs changeset version, which consumes the remaining .md files.
4. Appends the newly-consumed ids to <current-branch>.txt.
5. Deletes the hidden files (their consumption is already on record).

If changeset version fails, hidden files are renamed back to their original .md names so the working tree is left clean.

Release-PR branch naming

Releases land via a PR rather than a commit pushed straight to the target branch, so the branch the bump runs on is not the branch the release is for. Name the release PR branch release-pr/<target>, where <target> is the branch it merges into (main, release/11.1, …). The wrapper strips the release-pr/ prefix and keys the ledger by <target>, so every release for main accumulates in main.txt instead of scattering into a new file per PR.

The target branch is detected from git rev-parse --abbrev-ref HEAD (after stripping the prefix); a branch without the prefix is its own target. Set RELEASE_BRANCH to override (the prefix is stripped from it too).

Workflow requirement

Release branches must be merged back into main between releases, so that main sees the release branch's ledger entries before its own next release runs.