v2.1.0

What's Changed

• ci: publicar la verificacion como comentario del PR y fallar si algo falla by @JoshRob297 in #40
• docs: AGENTS.md (reglas del repo para agentes de IA) by @JoshRob297 in #41
• docs: Argon status and stale model reference cleanup by @JoshRob297 in #45
• docs: label the two illustrative examples as jsonc by @JoshRob297 in #46
• feat(models): optional backend-driven model discovery by @JoshRob297 in #47
• fix(models): make model discovery forward-only by @JoshRob297 in #48
• docs: document model discovery, and fix the tiered-id gate behind it by @JoshRob297 in #53
• feat(models): extend forward-only discovery to Claude by @JoshRob297 in #54
• fix(request): resolve the CLI client-signature version at startup by @JoshRob297 in #55
• build(deps-dev): bump the vitest group with 3 updates by @dependabot[bot] in #49
• build(deps-dev): bump @types/node from 26.6.4 to 26.6.5 by @dependabot[bot] in #50
• build(deps): bump zod from 4.6.5 to 4.6.7 by @dependabot[bot] in #51
• build(deps): bump @opencode-ai/plugin from 1.18.35 to 1.19.0 by @dependabot[bot] in #52
• docs(readme): add Spanish translation (README.es.md) by @JoshRob297 in #56
• docs(es): Spanish translations of the operational docs by @JoshRob297 in #57
• feat: ocultar los modelos de Google ajenos al catalogo curado by @JoshRob297 in #58

Full Changelog: v2.0.5...v2.1.0

v2.0.5 - Fix thinking effort on wrapped requests (#37)

v2.0.5 - Fix: thinking effort ignored on wrapped requests (#37)

Fixed

• Thinking variant (effort) ignored on wrapped requests (closes #37). In prepareAntigravityRequest, the backend model (gemini-3.8-flash-{low|medium|high}) was resolved from the name-derived tier before the variant/effort (providerOptions.google.thinkingLevel / generationConfig.thinkingConfig.thinkingLevel) was applied. As a result, wrapped requests — what OpenCode sends — always fell back to -low regardless of the selected effort. The variant is now extracted and applied before any backend-model resolution, for both wrapped and unwrapped bodies.

Case	Before	After
wrapped + effort high (providerOptions)	gemini-3.8-flash-low	gemini-3.8-flash-high
wrapped + effort high (generationConfig)	gemini-3.8-flash-low	gemini-3.8-flash-high
unwrapped + effort high	gemini-3.8-flash-high	gemini-3.8-flash-high

Tests

• 2 regression tests added. Full suite: 39/39 files, 1059 passed.

Credit

• Approach inspired by community PR #38 (@Atharva-G1205); landed as a minimal change that keeps the variants design.

What's Changed

• fix(request): apply thinking variant before backend model resolution (#37) by @JoshRob297 in #39

New Contributors

• @JoshRob297 made their first contribution in #39

Full Changelog: v2.0.4...v2.0.5

v2.0.4 - Prune dead consumer Gemini models

v2.0.4 - Prune dead consumer Gemini models

Removed

• antigravity-gemini-3.7-flash, antigravity-gemini-3.6-flash and antigravity-gemini-3.1-pro are now Google AI Ultra / Enterprise-only (plan change on 2026-10-13) and are no longer registered by the plugin. Selecting them previously failed with 404 NOT_FOUND.
• Same-day context: Gemini 4 ("Argon") released but restricted to Ultra / Enterprise; consumer plans are left with Gemini 3.8 Flash as the only Gemini model.

Changed

• Registered models are now exactly:
  • antigravity-gemini-3.8-flash (default)
  • antigravity-claude-sonnet-4-6
  • antigravity-claude-opus-4-6-thinking
  • antigravity-gpt-oss-120b-medium
• Whitelist and /antigravity-setup report updated to the trimmed catalog.
• Docs (README.md, docs/MODEL-VARIANTS.md) cleaned of dead model entries.
• Backend resolvers for 3.7/3.6/3.1 Pro are retained internally (harmless) but not surfaced.

Note

• Claude 4.6 + GPT-OSS remain until 2026-11-02, when they are removed by Google; after that consumer plans are left with Gemini 3.8 Flash only.
• Validated: tsc --noEmit clean, 39/39 test files (1057 passed).

v2.0.3 - Maintenance paused + dependency bumps

v2.0.3 - Maintenance paused + dependency bumps

Changed (docs)

• Maintenance paused pending a competitive Google model. Gemini 3.8 Flash is fast at code generation but still produces failures and does not reach parity with models such as DeepSeek, let alone frontier models.
• 2026-10-13: Google released Gemini 4 ("Argon") but restricted it to Google AI Ultra / Enterprise (not Google AI Pro). The same day, Gemini 3.7 Flash, 3.6 Flash and 3.1 Pro moved to Enterprise-only, leaving consumer plans with Gemini 3.8 Flash as the only Gemini model. Gemini 3.5 Pro has still not shipped.
• The Antigravity backend remains unreliable for paid plans (tier recognized but not applied), Claude 5.5 is unavailable, and Claude 4.6 + GPT-OSS-120b are removed on 2026-11-02.

Dependencies

• @opencode-ai/plugin 1.18.33 -> 1.18.35 (#36)
• @types/node 26.6.3 -> 26.6.4 (#35)
• vitest group (vitest, @vitest/coverage-v8, @vitest/ui) 5.0.2 -> 5.0.3 (#34)

Fixed

• README: removed a dead link to an internal entitlement report no longer published here.

No functional changes. The plugin stays usable as-is; pin this release for a stable docs + deps state.

What's Changed

• build(deps-dev): bump the vitest group with 3 updates by @dependabot[bot] in #34
• build(deps-dev): bump @types/node from 26.6.3 to 26.6.4 by @dependabot[bot] in #35
• build(deps): bump @opencode-ai/plugin from 1.18.33 to 1.18.35 by @dependabot[bot] in #36

Full Changelog: v2.0.2...v2.0.3

v2.0.2 - OpenCode v2 OAuth integration + API alignment

v2.0.2 - OpenCode v2 OAuth integration + API alignment

Fixes the OpenCode v2 authentication gap (issue #30) and aligns every v2 registration with the real @opencode/plugin 2.0.24 contracts.

Fixed

• OAuth integration (#30): setupV2 now registers the google Antigravity OAuth method via ctx.integration.transform (id: antigravity-oauth). opencode auth login and the TUI connect flow now offer OAuth with Google (Antigravity) instead of only the core API-key method.
• Multi-account pool: each successful authorization is persisted through persistAccountPool, appended/deduplicated into antigravity-accounts.json and picked up by the rotation engine. Connecting again adds the 4th, 5th, ... account.
• Model catalog: removed the non-existent ModelEditor.add branch; models register through editor.update("google", modelID, draft => ...), which seeds Model.Info.default and adds the model under an available provider. Drafts set enabled, capabilities and map variants to the real Model.Variant shape { id, settings }.
• Tools: ToolEditor.add no longer receives a bogus id (v2 derives it from name).
• Slash commands: execute now delivers output via ctx.session.synthetic; previously a returned string was silently ignored.
• Types: replaced the hand-rolled V2Context with typed local contracts mirroring @opencode/plugin@2.0.24 (no new runtime dependency).

Added

• Regression test asserting the v2 OAuth method is registered.

Upgrade

opencode auth login   # pick "OAuth with Google (Antigravity)"

Restart OpenCode so the new plugin build is loaded. Existing antigravity-accounts.json accounts are preserved.

Full Changelog: v2.0.1...v2.0.2

v2.0.1

What's Changed

• build(deps-dev): bump @types/node from 26.6.1 to 26.6.3 by @dependabot[bot] in #33
• build(deps): bump @opencode-ai/plugin from 1.18.31 to 1.18.33 by @dependabot[bot] in #32
• build(deps-dev): bump the vitest group with 3 updates by @dependabot[bot] in #31

Full Changelog: v2.0.0...v2.0.1

What's Changed

• build(deps-dev): bump @types/node from 26.6.1 to 26.6.3 by @dependabot[bot] in #33
• build(deps): bump @opencode-ai/plugin from 1.18.31 to 1.18.33 by @dependabot[bot] in #32
• build(deps-dev): bump the vitest group with 3 updates by @dependabot[bot] in #31

Full Changelog: v2.0.0...v2.0.1

What's Changed

• build(deps-dev): bump @types/node from 26.6.1 to 26.6.3 by @dependabot[bot] in #33
• build(deps): bump @opencode-ai/plugin from 1.18.31 to 1.18.33 by @dependabot[bot] in #32
• build(deps-dev): bump the vitest group with 3 updates by @dependabot[bot] in #31

Full Changelog: v2.0.0...v2.0.1

v2.0.0

Full Changelog: v1.19.0...v2.0.0

v1.19.0: Strict Tool Sanitization, Antigravity CLI 1.2.7 Parity & OpenCode v2 Dual Adapter

What's New in v1.19.0

Strict Tool Schema Sanitization (Fix HTTP 400 'property is not defined')

• Protobuf-backed Cloud Code API rejects tool calls when schema required arrays contain keys not defined in properties.
• Implemented recursive normalizer that prunes orphan required keys, omits degenerate empty required arrays, and enforces validation across pre-wrapped functionDeclarations tools.

Official Antigravity CLI v1.2.7 Signature Parity

• Sourced from reverse engineering Google's official agy v1.2.7 Linux x64 binary.
• Synchronized client signature headers and User-Agent: antigravity/cli/1.2.7 (aidev_client; os_type=linux; arch=amd64; cl=962369648; auth_method=consumer).
• Capped rate-limit retry backoffs to 30 seconds max for immediate multi-account rotation.

Model Catalog Addition

• Integrated Antigravity's native 120B model (antigravity-gpt-oss-120b-medium) with 131k context window and full multi-account routing support.

OpenCode v2 Dual-Compatibility Architecture

• Polymorphic entrypoint compatible simultaneously with OpenCode v1 (opencode-ai 1.18.x) and OpenCode v2 (@opencode/cli 2.0+).
• OpenCode v1 executes export default .server(ctx) returning { auth, event, tool }.
• OpenCode v2 executes export default .setup(ctx) dispatching native ctx.tool.transform, ctx.command.transform, ctx.model.transform, and native session hooks (http.request, http.response, retry).
• 1,048 automated tests in Vitest passing 100%.

Full Changelog: v1.18.0...v1.19.0

v1.18.0 — OPSEC Safety Shield & Account Rotation

🌟 What's New in v1.18.0

• 🛡️ OPSEC Safety Shield & Telemetry Engine (safety_shield):
  • Real-Time Extraction: Captures per-chunk safetyRatings probabilities (NEGLIGIBLE, LOW, MEDIUM, HIGH) across all harm categories from Gemini's streaming response.
  • OPSEC Notifications & Logs: Emits immediate warnings in debug logs and UI toasts when Google evaluates a response as HIGH harm or MEDIUM jailbreak.
  • Preventive Multi-Account Rotation (Account Shield): Automatically tracks consecutive high-risk responses per account (consecutiveHighRiskTriggers). When reaching the configured threshold (default 2), the plugin proactively rotates to the next available Google account (advanceToNextAccount) to prevent suspicious telemetry clustering on any single personal account.
• ⚙️ Configurable Settings: Controlled via ~/.config/opencode/antigravity.json:

  {
    "safety_level": "none",
    "safety_shield": {
      "enabled": true,
      "log_ratings": true,
      "show_toast": true,
      "auto_rotate_threshold": 2
    }
  }

• 🧪 1,046 passing tests across 38 test suites.

📦 Installation

Add this to your ~/.config/opencode/opencode.json:

{
  "plugin": ["github:JoshRob297/opencode-antigravity-auth"]
}

Full Changelog: v1.17.1...v1.18.0

Full Changelog: v1.17.1...v1.18.0

v1.17.1 — Strip x-goog-api-key & Fix 400 API_KEY_INVALID

🌟 What's New in v1.17.1

• 🔑 Strip x-goog-api-key Header (Fix 400 API_KEY_INVALID):
  • OpenCode automatically injects an AI Studio/Vertex API key via the x-goog-api-key header when intercepting requests to generativelanguage.googleapis.com.
  • The Antigravity backend (cloudcode-pa.googleapis.com) prioritized this header over Authorization: Bearer, rejecting requests with 400 INVALID_ARGUMENT (details[0].reason: API_KEY_INVALID, "API key not valid. Please pass a valid API key.").
  • Stripped x-goog-api-key alongside x-api-key in prepareAntigravityRequest, allowing requests to authenticate cleanly with OAuth Bearer tokens.
• 🧪 1,043 passing tests across 37 test suites.

📦 Installation

Add this to your ~/.config/opencode/opencode.json:

{
  "plugin": ["github:JoshRob297/opencode-antigravity-auth"]
}

Full Changelog: v1.17.0...v1.17.1

Full Changelog: v1.17.0...v1.17.1