[albrand]

Supersedes #290: the same change, rebased on ours at da8be25 (after the #296 merge and install 20) without a force-push. #290 stays open until this one merges, and its branch is kept.

Root cause

See #290. The rebase changes nothing in the PR's content.

Change

• fix/stale-draft-pencil (cc1efac) is rebased onto da8be25 as fix/stale-draft-pencil-on-da8be25 (10498b9, tree 53631f7e292e86b871cc874d922121a5245b8da8).
• The rebase had no conflicts. The PR's 3 commits map one to one. The four CLI thread-list commits it used to carry are already in ours.
• Diff equality: git diff da8be25...cc1efac52 | sha256sum = git diff da8be25 10498b93 | sha256sum = b9053b275918cc7e0364716629d9559a64ea5d3b2e1018259657136c6956326f (8 files, +290/−27).

Verification

Run at 10498b9 on the hsrpc worktree, after pnpm install --frozen-lockfile:

Check	Result
turbo run typecheck for app and thread-list	exit 0
thread-list tests	exit 0, 543 passed
app suite	first run 7,112 passed, 1 failed (ModelReasoningPicker.cold-open.test.tsx timed out); rerun exit 0, 7,113 passed, 17 skipped. The same suite at da8be25 passed 7,113/0. This PR doesn't touch the model picker.
cold-open picker test alone	5/5 passed at this head and 5/5 at da8be25. It's an intermittent timeout, carded separately; stamped reruns at both heads are in progress.
fork-features-check	99/99 fork features present
CI	run 40277376934, dispatched at this exact head

Owner walk: BLOCKED (browser input quarantine) and not run. It will be done after the adapter install. A Hermes clear plus green CI is enough to merge.

No issue: albrand/bb has issues disabled.

AGENT GENERATED