[orbitingflea]

This pull request resolves the merge of upstream 7ea6db4 into the fork: commit A (b87edc4) left conflict markers in 48 files. The fork now differs from upstream in 358 files, 4 fewer than before, and adds no lint suppression.

改了什么行为

Behaviour change: no. Fork users get upstream's changes, and each fork feature stays as it was. Where upstream's new code now decides, two things follow it. Three bare-link edge cases next to Chinese punctuation take the result of upstream's new pass. T3's new managed agy gets the fork's /compact support, as a user's own agy has it.

How

Each conflict keeps upstream's code and puts the fork's change back on top, with the fork's diff as small as the change allows. Fork copies of fixes that upstream now ships are gone, among them the fork's tracker of queued Claude steers. Claude usage and reset-credit reads follow upstream's new hub provider map.

vp run check:changed upstream/main passes on Node 26.12.1, and so does CI. Upstream now needs Node 26.12.1 and adds SQLite migrations 081 to 089, which cannot be undone, so a database backup comes before the first start.

Files

• cliproxyApi.ts: the fork's hooks sit in upstream's Claude and Codex hub entries (account label, the usage URL with reset credits, reading age, redeems through the usage source). Only redeems wait 30 s; upstream's new Antigravity read keeps 15 s.
• Antigravity: upstream runs every agy on T3's private profile. A user's own agy, named in Binary path, still runs on its own login and state. The runtime state is harness, managed or cli; /compact and the missing attachments folder apply to both agy runtimes.
• Claude steering: a steer still waits in Claude's queue. The fork's check now reads upstream's pending steers.
• Single tilde: upstream's user-message list (fix(web): preserve single tildes in sent user messages pingdotgg/t3code#35859) also covers review comments. The fork's port of open fix(web): preserve single tildes in agent replies pingdotgg/t3code#29270 stays for agent text, and upstream's timeline test now expects it.
• CJK links: upstream's pass runs first; the fork's test pins the three cases. Two differ from GFM at a space, an upstream behaviour.
• Dropped for upstream equivalents: the worktree cleanup check (fix(server): with newer Git, cleanup keeps worktrees whose repository is gone #50, upstream fix(server): git 2.56 broken-gitfile worktrees count as non-repositories again pingdotgg/t3code#33423), the Antigravity system-prompt check, duplicate Codex models lines.
• Citations: a Dashboard quote card does not navigate. History pages keep the fork's shown-only page before upstream's index.
• transferBudgetV2.integration.test.ts leaves out the fork's shownOnly opt-in. agentContextBudget.json: Claude and Codex text grow 56 bytes, all upstream's.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

ℹ️ No successful main baseline artifact is available yet. This run establishes the initial measurement.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	—	5.0 KiB	—	6.8 KiB	✅
Codex	Thread snapshot wire	—	3.8 KiB	—	4.9 KiB	✅
Codex	Live turn WebSocket wire	—	1.2 KiB	—	2.0 KiB	✅
Codex	Live turn WebSocket decoded	—	21.5 KiB	—	24.4 KiB	✅
Codex	Live turn messages	—	1	—	8	✅
Claude	Total thread wire	—	5.0 KiB	—	6.8 KiB	✅
Claude	Thread snapshot wire	—	3.8 KiB	—	4.9 KiB	✅
Claude	Live turn WebSocket wire	—	1.2 KiB	—	2.0 KiB	✅
Claude	Live turn WebSocket decoded	—	21.9 KiB	—	24.4 KiB	✅
Claude	Live turn messages	—	1	—	8	✅

Baseline: unavailable · PR result: c460138 · 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: 108.5 KiB
• Claude decoded thread snapshot: 108.8 KiB

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