chore(upstream): adopt the second 2026-08-21 T3 Code batch (#66)

[rynfar]

Upstream moved 23 commits past our cursor overnight, and one of them unblocked
work we had already deferred. This adopts 20 of the 23 and retires DEF-6.

DEF-6 is closed

#7089 (enable Cursor by default) was deferred yesterday because upstream's
own main was red on it: the commit added a test expecting every provider
enabled while the schema still had Grok and OpenCode off, contradicting an
existing test. #7725 reconciles them by deleting the contradictory test.
Re-probed exactly as the register instructed — both commits onto a scratch
branch, 39/39 pass in settings.test.ts, which is the only place in the repo
asserting a cursor default. Cursor is probed on fresh installs again; #7459's
substance survives because Grok and OpenCode stay false.

What else came in

Pinned-thread drag fixes (#7103, #7676) land on a Pylon feature. Sidebar
provider icons now resolve from the thread's own environment (#7292) —
default instance ids are driver slugs, so the old flat map collided across
environments, which our remote/multi-environment emphasis makes likelier to
hit. #7697 is a real security tightening: we accepted any vscode: or
cursor: URL for the OS handler, and it narrows that to a vscode-remote
host with an /ssh-remote+ path and no embedded credentials. #7561 moves the
preview loading bar off a rAF/React-state loop onto a compositor-driven CSS
animation. Plus slash-menu skills, terminal shifted-character encoding,
stream-following at the live edge, a longer git worktree add timeout, and
HTTPS-by-default GitHub clones.

Skipped three: #7283 (we already solved macOS lint gating better in #51, as
a path-filtered workflow rather than a gate job), #7728 and #7658 (fork
governance and upstream-specific prose).

The two that needed hands

#7286 CI parallelization could not be cherry-picked — our ci.yml had
diverged. Ported by hand: --parallel for non-server suites, three shards for
apps/server (278 files at fileParallelism: false), Rust into its own job.
Kept continue-on-error: true from #9 and added upstream's presence gate,
because they guard different failures — the gate stops the two shards that
legitimately produce nothing from racing for the artifact name, the flag stops
an exhausted org artifact quota from failing a run whose tests passed. Both
test jobs also pass --fail-if-no-match, which upstream does not use: without
it a filter matching no package only warns and exits 0, which is exactly the
silent no-op AGENTS.md warns about.

#7665 plans-out-of-repo is bigger than its title — it also adds a policy
doc, an AGENTS.md section, and config changes. Upstream deleted 32 files; we
tracked 34. The two survivors are live design docs for a shipping Pylon
provider, so under the policy this same commit adopts, they moved to
docs/internals/ rather than being deleted or stranded as tracked files
inside a now-gitignored directory.

Verification

• 19 test files / 374 tests across web, contracts, client-runtime, desktop,
  mobile, and the two touched server suites — all pass.
• Typecheck clean across 6 packages, each confirmed to have actually run.
• actionlint on the rewritten ci.yml: nothing beyond the pre-existing
  unknown Blacksmith runner labels.
• The sharded server command was run locally first: shard 1/3 took 95 of 278
  files, 900 passed. Its one failure is accountDrainEndToEnd.test.ts, which
  the previous batch's ledger already investigated as a local-only failure that
  reproduces on unmodified origin/pylon.

No UI screenshots: every visual change here is upstream's own, adopted
unmodified, and the composer/sidebar edits are structural rather than
appearance changes.

Written by Claude Opus 5 (1M context) in Claude Code.

---

^{Need help on this PR? Tag @codesmith with what you need. Autofix is disabled.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.4 KiB	13.4 KiB	−3 B (−0.0%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	−6 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.5 KiB	6.5 KiB	+3 B (+0.0%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.0 KiB	55.0 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	16	16	0 (0.0%)	21	✅
Claude	Total thread wire	13.5 KiB	13.5 KiB	+18 B (+0.1%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	+3 B (+0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.6 KiB	6.6 KiB	+15 B (+0.2%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	55.8 KiB	55.8 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	16	16	0 (0.0%)	21	✅

Baseline: 4f01bcf · PR result: 5c907dc · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.4 KiB
• Claude decoded thread snapshot: 110.0 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.