- Add OAuth handler with PKCE flow and local callback server
- Automatic project discovery via loadCodeAssist/onboardUser endpoints
- Store credentials with projectId for API calls
- Encode token+projectId as JSON for provider to decode
- Register as 'google-cloud-code-assist' OAuth provider
- Add new API type 'google-cloud-code-assist' for Gemini CLI / Antigravity auth
- Extract shared Google utilities to google-shared.ts
- Implement streaming provider for Cloud Code Assist endpoint
- Add 7 models: gemini-3-pro-high/low, gemini-3-flash, claude-sonnet/opus, gpt-oss
Models use OAuth authentication and have sh cost (uses Google account quota).
OAuth flow will be implemented in coding-agent in a follow-up.
- Add supportsXhigh() function to ai package for checking xhigh support
- Clamp xhigh to high for OpenAI models that don't support it
- Update coding-agent to use centralized supportsXhigh()
- gpt-5.2, gpt-5.2-codex now show xhigh in thinking selector
Closes#236
Add getApiKey hook to AgentLoopConfig that resolves API keys dynamically
before each LLM call. This allows short-lived OAuth tokens (e.g. GitHub
Copilot, Anthropic OAuth) to be refreshed between turns when tool
execution takes a long time.
Previously, the API key was resolved once when ProviderTransport.run()
was called and passed as a static string to the agent loop. If the loop
ran for longer than the token lifetime (e.g. 30 minutes for Copilot),
subsequent LLM calls would fail with expired token errors.
Changes:
- Add getApiKey hook to AgentLoopConfig (packages/ai)
- Call getApiKey before each LLM call in streamAssistantResponse
- Update ProviderTransport to pass getApiKey instead of static apiKey
- Update web-ui ProviderTransport with same pattern
Implement Agent Skills standard (https://agentskills.io/specification):
- Validate name (must match parent dir, lowercase, max 64 chars)
- Validate description (required, max 1024 chars)
- Warn on unknown frontmatter fields
- Warn on name collisions (keep first)
- Change prompt format to XML structure
- Remove {baseDir} placeholder (use relative paths)
- Add tests and update documentation
fixes#231
Previously, when using 'google-generative-ai' API with a custom baseUrl
in models.json, the baseUrl was ignored and requests always went to the
default Google endpoint.
Now the provider correctly passes model.baseUrl to the SDK's
httpOptions.baseUrl, enabling use of custom endpoints or API proxies.
Fixes#216
- Fix tool result format for Gemini 3 Flash Preview compatibility
- Use 'output' key for successful results (not 'result')
- Use 'error' key for error results (not 'isError')
- Per Google SDK documentation for FunctionResponse.response
- Improve type safety in google.ts provider
- Add ImageContent import and use proper type guards
- Replace 'as any' casts with proper typing
- Import and use Schema type for tool parameters
- Add proper typing for index deletion in error handler
- Add comprehensive test for Gemini 3 Flash tool calling
- Tests successful tool call and result handling
- Tests error tool result handling
- Verifies fix for issue #213Fixes#213
* use the correct Gemini 3 Flash Preview thinking levels
* fix a build error
* add changelog entry
* regenerate models
* make less assumptions about future models
Previous test used compressed 8k images (0.01MB) which was meaningless.
Now tests with actual large noise images that don't compress.
Realistic payload limits discovered:
- Anthropic: 6 x 3MB = ~18MB total (not 32MB as documented)
- OpenAI: 2 x 15MB = ~30MB total
- Gemini: 10 x 20MB = ~200MB total (very permissive)
- Mistral: 4 x 10MB = ~40MB total
- xAI: 1 x 20MB (strict request size limit)
- Groq: 5 x 5760px images (5 image + pixel limit)
- zAI: 2 x 15MB = ~30MB (50MB request limit)
- OpenRouter: 2 x 5MB = ~10MB total
Also fixed GEMINI_API_KEY env var (was GOOGLE_API_KEY).
Related to #120
Tested max 8kx8k images per provider:
- Anthropic: 100 (explicit limit, fails at 101)
- OpenAI: 100-200 (100 works, 200 times out)
- Mistral: 8 (explicit limit, fails at 9)
- xAI: 100-150 (100 works, 150 times out)
- Groq: 0 (8k exceeds 33M pixel limit)
- zAI: 400 (context window limited at 500)
- OpenRouter: 40 (context window limited at 50)
- Gemini: untested (no API key in test env)
Key finding: Anthropic's 'many images' rule does NOT cause API errors.
100 x 8kx8k images work fine. Anthropic likely auto-resizes internally.
Related to #120
Tests max image count, size, dimensions, and 8k stress test for:
- Anthropic, OpenAI, Gemini, Mistral, OpenRouter, xAI, Groq, zAI
Key finding: Anthropic's 'many images' rule (>20 images = 2000px max)
does NOT cause API errors. 100 x 8k images work fine. Anthropic likely
auto-resizes internally.
Related to #120
Tests provider-specific image limitations across all supported providers:
- Maximum number of images in context
- Maximum image size (bytes)
- Maximum image dimensions
Discovered limits (Dec 2025):
- Anthropic: 100 images, 5MB per image, 8000px max dimension
- OpenAI: 500 images, >=25MB per image
- Gemini: ~2500 images, >=40MB per image
- Mistral: 8 images, ~15MB per image
- OpenRouter: ~40 images (context limited), ~15MB per image
- Add AgentToolUpdateCallback type and optional onUpdate callback to AgentTool.execute()
- Add tool_execution_update event with toolCallId, toolName, args, partialResult
- Normalize tool_execution_end to always use AgentToolResult (no more string fallback)
- Bash tool streams truncated rolling buffer output during execution
- ToolExecutionComponent shows last N lines when collapsed (not first N)
- Interactive mode handles tool_execution_update events
- Update RPC docs and ai/agent READMEs
fixes#44
Previously, when reasoning was not specified, some providers like Gemini
with 'dynamic thinking' enabled by default would still use thinking.
Now explicitly sets thinkingEnabled: false (Anthropic) and
thinking: { enabled: false } (Google) when reasoning is undefined.
Closes#180
