release: 1.2.0-rc.6

[fujibee]

Declared reviewers: 1

Landing on integration/remote. Head 5b43ddf225219fcb486b7db30c510d9b21231134.

What this is

VERSION → 1.2.0-rc.6, and the two files derived from it, written by scripts/release/sync-version.sh rather than by hand:

VERSION                     1.2.0-rc.5 -> 1.2.0-rc.6
package.json                1.2.0-rc.5 -> 1.2.0-rc.6
.claude-plugin/plugin.json  1.2.0-rc.5 -> 1.2.0-rc.6

scripts/release/sync-version.sh --check passes at this head — the same check CI runs as the pre-release guard, so a forgotten sync fails there rather than shipping a package whose package.json disagrees with VERSION.

Three files, matching release: 1.2.0-rc.5 exactly. No CHANGELOG.md entry: the rc line has not carried one, and inventing a format here would make the next release's diff ambiguous.

What is in this rc that was not in rc.5

The two that change behaviour a user can feel:

• A connected team's sync engine is started when an agent turns up (A reboot leaves every sync engine dead: start one when an agent joins a connected team #774). Before, a restart left every engine dead and nothing restarted one — send kept succeeding locally and nothing arrived, so the only symptom was silence.
• A refusal from the server no longer kills the engine, and can be read (A refusal the operator could act on arrives as a transport failure, and the sender is told nothing #773). It records the status, code and host verbatim, backs off, and keeps the loop; status reports it, and reports nothing once a later cycle succeeds.

And one that shows up as CI stability rather than as behaviour:

• The pidfile handover no longer loses the successor's record (bats suite flakes across the process-lifecycle tests: main itself fails ~half its runs, a different test each time #595). A departing watcher could delete a pidfile its successor had just written, because the successor signalled it before claiming the slot. That is the failure that had bats (ubuntu-latest 4/4) red across every open PR.

Not claimed here

That #595 is closed. One branch has been rebased onto the fix and passed that shard — one sample, on an assertion that fails intermittently. The issue stays open until more of the queue has crossed it.

Measurements, at this head

scripts/release/sync-version.sh --check   passes
git diff --stat                           3 files, +3 -3

No test run is quoted: this changes three version strings and no code. The suite result that matters is the one on the head this branches from, 5609129e89eaecbb10fbd0b536dc9a757f2071d2, which carries #775 and #803.

Drift

merge-base                 5609129e89eaecbb10fbd0b536dc9a757f2071d2
origin/integration/remote  5609129e89eaecbb10fbd0b536dc9a757f2071d2
git diff --stat <merge-base> origin/integration/remote -- VERSION package.json .claude-plugin/plugin.json
  → empty

Re-measured immediately before landing.