Using the Antigravity agy CLI as the custom binary broke most threads. T3 sent the managed runtime's yolo (Full access) and auto_edit (Auto-accept edits) session modes. The CLI only knows default, accept-edits, and plan, so the session failed to open:
| 1 | Invalid value "yolo" for session config option "mode": expected one of default, accept-edits, plan |
Full access is the default mode, so every new CLI thread failed. See #32856.
When the runtime is the CLI, T3 sends default in place of yolo and auto_edit. In default the CLI asks before it acts, and T3 already answers those requests from the live permission setting: Full access allows, Auto-accept edits allows edits, and Supervised asks. So permission changes apply on the next turn, with no bypass flag and no process restart. The managed runtime is unchanged.
This takes over #32875 by @Q3yp, and their commits are kept. That PR launched agy acp --dangerously-skip-permissions for Full access and replaced the session on every permission change. Review found that a downgrade made while a session was still opening could leave the bypass flag on. Answering permissions in T3 removes the flag, and with it the race, and cuts the change to about 13 source lines.
agy 1.5.3 against a T3 dev server. On main, a Full access thread fails with the error above. On this branch one agy acp process (same PID, no flag) handles every switch:default. It fails on main with ["yolo"].Rounds: 4
| Delegate | Round 1 | Round 2 | Round 3 | Round 4 |
|---|---|---|---|---|
| Claude Opus 5.5 | 8/10 | 9/10 | 9/10 | 10/10 |
| Muse Wonder Full | 8/10 | 9/10 | 9/10 | 9/10 |
| Codex GPT-6.1 Sol | 7/10 | 9/10 | 9/10 | 9/10 |
Round 1 reviewed the bypass-flag design and led to the simpler fix above.
Created with Claude Opus 5.5 in Claude Code (T3 Code).
š¤ Generated with Claude Code
Added 4 commits
Verdict: Approved at 64b56d6
Macroscope's review found this PR approvable ā This adapts Antigravity CLI mode requests so existing Full access and Auto-accept edits selections can open sessions. Product permission defaults and managed-harness behavior remain unchanged.
You can add or adjust custom eligibility rules. Learn more.
ā Thread transfer remains within every enforced ceiling.
| Provider | Metric | Main baseline | This PR | Impact | PR ceiling | |
|---|---|---|---|---|---|---|
| Codex | Total thread wire | 5.0 KiB | 5.0 KiB | 0 B (0.0%) | 6.8 KiB | ā |
| Codex | Thread snapshot wire | 3.8 KiB | 3.8 KiB | 0 B (0.0%) | 4.9 KiB | ā |
| Codex | Live turn WebSocket wire | 1.2 KiB | 1.2 KiB | 0 B (0.0%) | 2.0 KiB | ā |
| Codex | Live turn WebSocket decoded | 21.5 KiB | 21.5 KiB | 0 B (0.0%) | 24.4 KiB | ā |
| Codex | Live turn messages | 1 | 1 | 0 (0.0%) | 8 | ā |
| Claude | Total thread wire | 5.0 KiB | 5.0 KiB | 0 B (0.0%) | 6.8 KiB | ā |
| Claude | Thread snapshot wire | 3.8 KiB | 3.8 KiB | 0 B (0.0%) | 4.9 KiB | ā |
| Claude | Live turn WebSocket wire | 1.2 KiB | 1.2 KiB | 0 B (0.0%) | 2.0 KiB | ā |
| Claude | Live turn WebSocket decoded | 21.9 KiB | 21.9 KiB | 0 B (0.0%) | 24.4 KiB | ā |
| Claude | Live turn messages | 1 | 1 | 0 (0.0%) | 8 | ā |
Baseline: b24959d Ā· PR result: 64b56d6 Ā· Source CI: success
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.
Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.
