Skip to content

fix(antigravity): let max and ultra select Hydrogen's xhigh tier and record its CCA verification - #8644

Merged
lidge-jun merged 5 commits into
lidge-jun:devfrom
andrew05060414:fix/antigravity-hydrogen-xhigh-max
Nov 13, 2026
Merged

lidge-jun merged 5 commits into
lidge-jun:devfrom
andrew05060414:fix/antigravity-hydrogen-xhigh-max

Conversation

@andrew05060414

@andrew05060414 andrew05060414 commented Nov 13, 2026 •

Copy link
Copy Markdown
Contributor

Summary

Follow-up to #8636, which folded Gemini 4 Hydrogen's discovered -xhigh tier into its family. Two gaps remain, both small:

  • max and ultra still clamp to high on a four-tier family. discoveredAntigravityEffortWireModelId selects the xhigh tier only for the literal xhigh; max and ultra go through resolveAntigravityThinkingLevel and land on gemini-4-hydrogen-high. A client that asks for its top effort under either name therefore gets one rung below what the family offers. This PR makes xhigh, max and ultra share one set (ANTIGRAVITY_ABOVE_HIGH_EFFORTS): they select a discovered xhigh tier when the family has one, and clamp to high exactly as before when it does not. Argon, Carbon and every other three-tier family are unchanged, and the existing three-tier test now asserts that for all three efforts.
  • Hydrogen on Cloud Code Assist is no longer unobserved. feat(antigravity): route Gemini 4 Hydrogen's discovered xhigh tier through the CLI User-Agent #8636 documented Hydrogen on CCA as anticipated but not seen. On 2026-11-12 an entitled account's CCA discovery listed exactly gemini-4-hydrogen-low/-medium/-high/-xhigh, and an xhigh request through this proxy answered 200 (details under Verification). The comments in src/adapters/google-antigravity-wire.ts and src/providers/antigravity-models.ts, the Hydrogen test header, structure/providers-and-adapters.md and the English providers guide now say so. The guide also notes that max/ultra select xhigh.

No change to discovery, the snapshot format, the User-Agent rule, the Hydrogen high default, or partial-family handling.

Verification

Failing first. The test commit (test(antigravity): pin max and ultra ...) extends tests/adapters/google/gemini-4-hydrogen-antigravity.test.ts. The same resolution run against the dev source and against this branch:

before: xhigh->gemini-4-hydrogen-xhigh max->gemini-4-hydrogen-high  ultra->gemini-4-hydrogen-high  high->gemini-4-hydrogen-high undefined->gemini-4-hydrogen-high
after:  xhigh->gemini-4-hydrogen-xhigh max->gemini-4-hydrogen-xhigh ultra->gemini-4-hydrogen-xhigh high->gemini-4-hydrogen-high undefined->gemini-4-hydrogen-high

Local checks on this branch (based on dev at 61dfb8be4):

  • bun run typecheck: exit 0, "No type errors in 2,492 files".
  • bun run structure:check: exit 0, "structure/ SSOT checks passed".
  • bun run privacy:scan: exit 0, "Privacy scan passed".
  • bun test tests/adapters/google/gemini-4-hydrogen-antigravity.test.ts tests/adapters/google/antigravity-discovered-families.test.ts tests/adapters/google/gemini-4-argon-antigravity.test.ts tests/providers/provider-antigravity-effort-families.test.ts tests/providers/provider-antigravity-wire-snapshot.test.ts tests/providers/provider-antigravity-family-catalog.test.ts tests/adapters/google/google-antigravity-wire.test.ts tests/ci-workflows/repo-hygiene.test.ts: exit 0, 144 pass, 0 fail across 8 files. The existing three-tier tests (Argon xhigh → -high, partial families keep their wire identity) are unchanged and pass.

Resource exception: bun run test:changed and the full bun run test were not completed. Several worktrees on this machine were running suites at once and share the user-scoped test-run lock; the test:changed attempt waited 807 s for the lock and was then terminated by the 900 s suite limit (exit 124) with 70 failures, none in an Antigravity, Google or catalog test file, most of them 5 s timeouts. That run was on the earlier, larger version of this change before rebasing. Import-graph coverage beyond the files listed above is left to CI.

Live check against Cloud Code Assist (2026-11-12, local test port, temporary OPENCODEX_HOME holding one Antigravity account and only the google-antigravity provider, deleted afterwards; the build was this change's earlier version, whose xhigh routing matches dev plus this PR):

  • One GET /v1/models, i.e. one discovery: google-antigravity/gemini-4-hydrogen appeared once with efforts low, medium, high, xhigh, and no separate gemini-4-hydrogen-xhigh row. The saved wire snapshot recorded xhigh: "gemini-4-hydrogen-xhigh" from CCA's own listing.
  • One POST /v1/responses, model: google-antigravity/gemini-4-hydrogen, reasoning.effort: xhigh: HTTP 200, status: completed, 25 reasoning tokens.
  • The proxy does not log the outbound wire id, so the same request was rebuilt offline from that snapshot with the Google adapter: xhigh and max produced model: gemini-4-hydrogen-xhigh with the CLI User-Agent and thinkingConfig: {includeThoughts: true} (no thinkingLevel); high produced gemini-4-hydrogen-high.

Checklist

  • Scope stays focused and avoids unrelated cleanup.
  • Docs or release notes were updated when needed.
  • Security-sensitive changes were reviewed for secrets, auth, and unsafe defaults.

Maintainer follow-up (head 45fdd2a)

  • Spanish guide (es/guides/providers.md) mirrors the English sentence: Hydrogen's CCA status and the max/ultra rule.
  • Stale comment in antigravity-models.ts (rule 2/3) now says discovery clamps above-high efforts only for a family without an xhigh tier.
  • New tests: adapter requests for xhigh/max/ultra/high/no effort, with and without the catalog ladder; routing from a restored persisted snapshot (four-tier Hydrogen selects -xhigh, three-tier Carbon clamps to -high). Against dev's source, the three xhigh-routing tests fail; with this PR, 9 of 9 pass.
  • Merged latest dev. bun run typecheck, structure:check and privacy:scan pass, and these files pass with 0 failures, one process each: gemini-4-hydrogen-antigravity (9), gemini-4-argon-antigravity (19), antigravity-discovered-families (11), google-antigravity-wire (62), antigravity-routing-carry (28), antigravity-account-roster (14), provider-antigravity-wire-snapshot (6), provider-antigravity-effort-families (19), docs-spanish-localization (30), file-size-ratchet (9), repo-hygiene (16).

Review readiness checklist

This PR stays in draft until every box below is ticked. Tick all four boxes once the requirements are met:

  • Required local validation passed; commands, results, and any full-suite exception are documented.

  • I pushed my PR to a recent dev commit (at most 10 behind; a maintainer may still ask for the exact tip before merge).

  • I resolved all correct Codex and CodeRabbit findings.

  • My PR is ready for review.

…record its CCA verification

On a discovered family that has an xhigh tier, requested max and ultra now
select it instead of clamping to high; three-tier families are unchanged.
Comments and docs that called Hydrogen on CCA unobserved now record the
2026-11-12 check: an entitled account's CCA discovery listed the four tiers
and gemini-4-hydrogen-xhigh answered 200 under the CLI User-Agent.
@coderabbitai

coderabbitai Bot commented Nov 13, 2026 •

Copy link
Copy Markdown
Contributor

Explain Change →

Caution

Review failed

The pull request is closed.

  • Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

Share

@github-actions

Copy link
Copy Markdown
Contributor

✅ Deterministic PR hygiene checks passed.

@github-actions github-actions Bot added the bug Something isn't working label Nov 13, 2026
@github-actions

github-actions Bot commented Nov 13, 2026 •

Copy link
Copy Markdown
Contributor

✅ READY

  • all PR quality gates passed; the review readiness checklist is complete.

Review readiness checklist

  • ✅ Required local validation passed; commands, results, and any full-suite exception are documented.
  • ✅ I pushed my PR to a recent dev commit (at most 10 behind; a maintainer may still ask for the exact tip before merge).
  • ✅ I resolved all correct Codex and CodeRabbit findings.
  • ✅ My PR is ready for review.

✅ 4/4 boxes ticked.

This pull request has been marked Ready for Review.
The review-ready label marks this PR as ready; review automation runs independently.
Maintainers notified: @lidge-jun @Ingwannu

Hygiene

✅ Deterministic PR hygiene checks passed.

@lidge-jun

Copy link
Copy Markdown
Owner

Maintainer judgment

Merging at head 45fdd2a9743f531816e78bb26e4c340b74d279eb.

  • Why: after feat(antigravity): route Gemini 4 Hydrogen's discovered xhigh tier through the CLI User-Agent #8636, a client asking a four-tier Hydrogen family for its top effort as max or ultra got -high, one rung below what the family offers. Now xhigh, max and ultra share one rule. They select a discovered xhigh tier when the family has one, and clamp to high exactly as before when it does not. Argon, Carbon, 3.8 Flash and Claude 5.5 are unchanged, and the static path and Claude 4.6's thinking-level max are untouched.
  • Evidence: Hydrogen on CCA rests on the contributor's live check on 2026-11-12: an entitled account's discovery listed exactly gemini-4-hydrogen-low/-medium/-high/-xhigh, and an xhigh request returned 200. No account here is entitled to Hydrogen, so I did not reproduce it. Hydrogen still appears only when an account's own discovery lists it, so an account that does not list it is unaffected.
  • Reviews on this exact head: a fresh independent review and a separate security review (accounts, client identity, host, snapshot validation, disclosure) both returned APPROVE (gpt-6.1-terra, high). Earlier non-blocking notes on a stale comment and missing adapter and restore-path coverage were fixed before this head.
  • Declined (non-blocking): the reviewer suggested qualifying "requesting max or ultra also selects xhigh" in the guides with "when the family has an xhigh tier". The sentence describes Hydrogen as listed with its four levels, xhigh included, and the structure doc states the general condition, so the guides stay as written.
  • CI: no job on this head has failed. The remaining jobs (test changed, gates, docker smoke, docs build, structure gate, npm-global, storage policy, API usage) were still running at merge time; per the maintainer instruction, the merge does not wait for them.
  • Merge path: squash with --match-head-commit; --admin is used only if the one-approval rule is the only block.

@github-actions
github-actions Bot marked this pull request as ready for review November 13, 2026 06:02
@lidge-jun
lidge-jun merged commit 585f048 into lidge-jun:dev Nov 13, 2026
29 checks passed
lidge-jun added a commit that referenced this pull request Nov 13, 2026
Brings in dev through 6f58f62, including #8612 (recover a
force-revoked ephemeral Child by its link id: tombstones, ocx link
recover), #8635 and #8644.

Conflict resolution in removeReserved keeps both sides: dev's tombstone
is written before the first intent for a forced revoke of an ephemeral
Child, with this PR's cleanup condition (cleanup already recorded by
the owner's journal is not rerun); every way out that does not commit
(claim failure, a changed record, a failed key revoke or journal
intent, the post-commit catch) drops the force's own tombstone, which
dev's helper does only while the link still exists; a committed
non-force removal settles earlier tombstones of the link. remove()
admits the alias lock for a tombstoned alias using the removal
target's record. The generated skill surface was regenerated.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

bug Something isn't working review-ready

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants