Release v2.0.4

Release Notes v2.0.4

Released: 2026-01-09

This release focuses on credential management and provider flexibility. API keys can now be stored in git-iris configuration instead of relying solely on environment variables, and the parallel analysis tool gains full Google/Gemini support with improved error handling.

🌟 Highlights

🔐 Config-Based API Key Resolution

API keys can now be stored directly in your .irisconfig file instead of requiring environment variables. The resolution order is: config → environment variable → client default. This improves portability for containerized deployments and IDE-integrated workflows.

✨ Google/Gemini Support for Parallel Analysis

The parallel_analyze tool now supports Google/Gemini as a provider alongside OpenAI and Anthropic. All three providers use consistent API key resolution and error handling patterns.

🔒 API Key Format Validation

New validation catches common API key misconfigurations early. The system checks expected prefixes (sk- or sk-proj- for OpenAI, sk-ant- for Anthropic) and minimum length requirements before making API calls.

🛡️ Sanitized Error Messages

Error messages during authentication failures no longer risk exposing API key material. Client creation errors return generic messages without revealing key prefixes or partial credentials.

🔧 Provider Infrastructure

• Added resolve_api_key() function in src/agents/provider.rs with ApiKeySource enum to track where keys originate (Config, Environment, ClientDefault)
• Provider builder functions (openai_builder, anthropic_builder, gemini_builder) now accept optional api_key parameter and return Result
• Added validate_api_key_format() method to Provider for format validation with provider-specific prefix checking
• OpenAI now accepts both sk- and sk-proj- prefixes, supporting project-scoped API keys
• Added api_key_if_set() helper to ProviderConfig that returns None for empty strings

🤖 Agent Framework

• IrisAgent gains get_api_key() method to retrieve provider API keys from stored config
• IrisAgentService exposes api_key() accessor for config-based key retrieval
• StatusMessageGenerator constructor now accepts optional API key parameter
• All agent builder paths thread API keys through to subagents and debug agents

⚡ Parallel Analysis

• SubagentRunner enum extended with Gemini variant for Google provider support
• Added resolve_openai_client(), resolve_anthropic_client(), and resolve_gemini_client() helper methods with consistent resolution logic
• Removed silent fallback to OpenAI when requested provider fails — users now receive clear error messages indicating which provider failed
• ParallelAnalyze::new() and with_timeout() accept explicit api_key parameter

🐛 Fixes

• Fixed error handling in LLM builder functions — replaced panics with context-aware error returns using anyhow
• src/studio/app/mod.rs now passes API key from agent config to StatusMessageGenerator

Breaking Changes

• Provider builder return type changed — openai_builder(), anthropic_builder(), and gemini_builder() now return Result<...Builder> instead of the builder directly. Callers must handle the error case with ? or explicit error handling.

• ParallelAnalyze constructor signature changed — ParallelAnalyze::new() and with_timeout() now require an additional api_key: Option<&str> parameter.

• No silent provider fallback — Previously, if the configured provider (e.g., Anthropic) failed to initialize, the system would silently fall back to OpenAI. This release removes that behavior. If your configured provider fails, you'll receive an explicit error message.

Upgrade Notes

• If you prefer config-based API key storage, add your key to .irisconfig:

  git-iris config --api-key YOUR_API_KEY

  The key will be stored under the current provider's configuration.

• Environment variables continue to work as before. Config keys take precedence when both are set.

• If you were relying on silent fallback to OpenAI when another provider failed, ensure your provider credentials are valid or explicitly configure OpenAI as your provider.