Solution for long term memory for agent coding CLIs and to facilitate handoff between different agent vendors
Rust
fix(changelog): stop entries landing in already-released sections
AkitaOnRailsec9bdaae15fd
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.ymlindex bcd73d25..5940c538 100644--- a/.github/workflows/ci.yml+++ b/.github/workflows/ci.yml@@ -145,7 +145,13 @@ jobs:runs-on: ubuntu-lateststeps:- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1+ with:+ # `check-changelog-frozen.sh` diffs against the merge base, so it+ # needs history rather than a shallow clone.+ fetch-depth: 0- run: scripts/check-changelog-sections.sh+ - name: released sections stay frozen+ run: scripts/check-changelog-frozen.sh "origin/${{ github.base_ref || github.event.repository.default_branch }}"# End-to-end docker smoke: packages the release-build binary and confirms# the image runs. Catches Dockerfile drift (missing COPY, bad ENTRYPOINT)diff --git a/CHANGELOG.md b/CHANGELOG.mdindex f7b79708..347341cb 100644--- a/CHANGELOG.md+++ b/CHANGELOG.md@@ -7,6 +7,34 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0## [Unreleased]+### Fixed+- CI now rejects a change that writes into an already-released CHANGELOG+ section. `bin/release` renames `## [Unreleased]` to `## [X.Y.Z]`, so a branch+ opened before a release carries a diff anchored at the old line numbers and+ git merges it into whatever section now occupies them — silently, with no+ conflict. Three entries landed in the wrong release this way (#491 into+ 1.32.0, #502 into 1.32.2, #517 into 1.33.0), each claiming a fix shipped in a+ version that did not contain it while the version that did listed nothing.+ `scripts/check-changelog-frozen.sh` compares a branch against its merge base+ and fails on any touched line below `[Unreleased]`.+- The `ingest (server, this process)` counters in `ai-memory status` now move+ for events delivered over `POST /hook/batch`. They were instrumented only in+ the per-event `POST /hook` handler, while the spool drain posts batches and+ falls back to `/hook` only against a pre-upgrade server, so a current client+ against a current server left the whole section reading `accepted 0`,+ `last write: -` and zero sheds while observations were landing normally — the+ section exists precisely to tell "hooks are not arriving" apart from "hooks+ are arriving but nothing is stored", and it reported the same thing for both.+ A batch 429 also now counts every item it rejects, not just the one that+ found no permit, so shed volume is comparable between the two routes instead+ of understated by up to the batch size (#516).+- `last write` in the same section now advances only when an event actually+ cleared the writer. `handle_hook` stamped it unconditionally after processing+ returned, and processing swallows its errors, so a store rejecting every+ event — read-only database, exhausted disk, a failed migration — still+ reported a fresh write time and looked healthy (#516).++## [1.33.0] - 2026-08-28### Added@@ -38,23 +66,6 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0HTTP 400 and disappear from memory while ASCII events still worked. A nativeWindows loopback test now round-trips Chinese and Portuguese text byte forbyte (#500).-- The `ingest (server, this process)` counters in `ai-memory status` now move- for events delivered over `POST /hook/batch`. They were instrumented only in- the per-event `POST /hook` handler, while the spool drain posts batches and- falls back to `/hook` only against a pre-upgrade server, so a current client- against a current server left the whole section reading `accepted 0`,- `last write: -` and zero sheds while observations were landing normally — the- section exists precisely to tell "hooks are not arriving" apart from "hooks- are arriving but nothing is stored", and it reported the same thing for both.- A batch 429 also now counts every item it rejects, not just the one that- found no permit, so shed volume is comparable between the two routes instead- of understated by up to the batch size (#516).-- `last write` in the same section now advances only when an event actually- cleared the writer. `handle_hook` stamped it unconditionally after processing- returned, and processing swallows its errors, so a store rejecting every- event — read-only database, exhausted disk, a failed migration — still- reported a fresh write time and looked healthy (#516).-## [1.32.2] - 2026-08-26### Fixeddiff --git a/scripts/check-changelog-frozen.sh b/scripts/check-changelog-frozen.shnew file mode 100755index 00000000..7af0ecb3--- /dev/null+++ b/scripts/check-changelog-frozen.sh@@ -0,0 +1,78 @@+#!/usr/bin/env bash+# Fail if a change adds or edits entries inside an already-released+# CHANGELOG section.+#+# `bin/release` renames `## [Unreleased]` to `## [X.Y.Z]`. A branch opened+# before that release still carries a CHANGELOG diff anchored at the old+# `[Unreleased]` line numbers, and git merges it into whatever section now+# occupies them — no conflict, no warning. The entry then claims a change+# shipped in a release that did not contain it, and the release that *did*+# ship it lists nothing.+#+# Three times in one week: #491 into [1.32.0], #502 into [1.32.2], #517 into+# [1.33.0]. Each found by hand, twice only because an unrelated merge+# conflict forced someone to look at the file.+#+# Usage: check-changelog-frozen.sh [base-ref] (default: origin/main)+#+# Compares the merge base of <base-ref> and HEAD against HEAD, and rejects+# any changed CHANGELOG line that sits below the `## [Unreleased]` heading.+# Deliberate historical corrections on the default branch are unaffected:+# this asks what a branch changes relative to where it forked, not what the+# file looks like versus a tag.++set -euo pipefail++REPO_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"+cd "$REPO_ROOT"++BASE_REF="${1:-origin/main}"+git rev-parse --verify "$BASE_REF" >/dev/null 2>&1 || {+ echo "check-changelog-frozen: base ref '$BASE_REF' not found; skipping."+ exit 0+}++MERGE_BASE="$(git merge-base "$BASE_REF" HEAD)"+if [[ "$MERGE_BASE" == "$(git rev-parse HEAD)" ]]; then+ echo "check-changelog-frozen: HEAD is an ancestor of $BASE_REF; nothing to check."+ exit 0+fi++# First line number of the first released section in the *new* file.+FROZEN_FROM="$(awk '/^## \[[0-9]/{print NR; exit}' CHANGELOG.md)"+if [[ -z "$FROZEN_FROM" ]]; then+ echo "check-changelog-frozen: no released section yet."+ exit 0+fi++# Every new-file line number this diff touches, via the unified hunk headers.+TOUCHED="$(git diff --unified=0 "$MERGE_BASE" HEAD -- CHANGELOG.md \+ | awk '/^@@/{ split($3, a, ","); start = a[1] + 0; if (start < 0) start = -start;+ count = (a[2] == "" ? 1 : a[2] + 0);+ for (i = 0; i < count; i++) print start + i }')"++BAD=""+for ln in $TOUCHED; do+ if (( ln >= FROZEN_FROM )); then+ BAD+=" line $ln: $(sed -n "${ln}p" CHANGELOG.md)"$'\n'+ fi+done++if [[ -z "$BAD" ]]; then+ echo "CHANGELOG: no released section was modified."+ exit 0+fi++echo "error: this change modifies an already-released CHANGELOG section." >&2+echo >&2+printf '%s' "$BAD" >&2+cat >&2 <<'EOF'++Released sections are frozen. An entry for unreleased work belongs under+`## [Unreleased]`.++This usually means the branch was opened before a release was cut, and git+merged the entry into the section that now occupies those lines. Move it up+to `[Unreleased]`; the content does not change, only the section.+EOF+exit 1