Implement critical production readiness features for AI integration: per-request provider config, admin dashboard, workflow inspection, and plugin framework foundation.
Per-Request Provider Config:
- Add ProviderConfig struct to all AI request types
- Update queue to resolve provider/model/API key from cascade (repo > org > system)
- Pass resolved config to AI sidecar on every request
- Fixes multi-tenant issue where all orgs shared sidecar's hardcoded config
Admin AI Dashboard:
- Add /admin/ai page with sidecar health status
- Display global operation stats (total, 24h, success/fail/escalated counts)
- Show operations by tier, top 5 repos, token usage
- Recent operations table with repo, operation, status, duration
- Add GetGlobalOperationStats model method
Workflow Inspection:
- Add InspectWorkflow client method and types
- Implement workflow-inspect queue handler
- Add notifier trigger on workflow file push
- Analyzes YAML for syntax errors, security issues, best practices
- Returns structured issues with line numbers and suggested fixes
Plugin Framework (Phase 5 Foundation):
- Add external plugin config loading from app.ini
- Define ExternalPlugin interface and manager
- Add plugin.proto contract (Initialize, Shutdown, HealthCheck, OnEvent, HandleHTTP)
- Implement health monitoring with auto-restart for managed plugins
- Add event routing to subscribed plugins
- HTTP proxy support for plugin-served routes
This completes Tasks 1-4 from the production readiness plan and establishes the foundation for managed plugin lifecycle.
Refactor AI service layer to reduce code duplication and improve consistency.
Changes:
- Rename AIOperationRequest to OperationRequest for consistency
- Extract shared logic for issue-targeted operations (respond, triage) into triggerIssueAIOp helper
- Standardize field alignment in struct definitions
- Remove redundant error handling patterns
This reduces the API operations file by ~40 lines while maintaining identical functionality.
Implement core AI service infrastructure including agent operations, escalation handling, and asynchronous queue processing.
New services:
- Agent service: Handles Tier 2 AI operations with action runner integration
- Queue service: Asynchronous processing of AI operations with retry logic
- Escalation service: Routes complex issues to staff with configurable rules
- Notifier service: Sends notifications for AI operation results
Additional changes:
- Add GitCaddy AI system user (ID: -3) for bot operations
- Add AIConfig to repository units
- Add AI-specific error codes (rate limiting, service errors, etc.)
- Extend AI client with GenerateIssueResponse method
- Add AISettingsV2 struct for repository-level AI configuration
The queue system enables non-blocking AI operations with proper error handling and rate limiting.
