You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Gate A2A agent execution behind per-agent user consent. (#492)
Replace loosely typed federationMetadata dictionaries with provider-specific Pydantic models for AWS AgentCore MCP, AWS AgentCore A2A, and Azure AI Foundry A2A. (#490)
Restrict direct-connect A2A routes to reject DCR clients before agent lookup and prevent unauthorized invocation. (#495)
Validate non-gRPC transports for A2A agents to ensure only supported transport protocols are used. (#498)
Add approval gate toolbar and active run polling to provide real-time status updates during workflow execution. (#493)
Allow authorized users to trigger runs in view-only mode. (#500)
Serve public managed Agent Card endpoints with Registry proxy URLs, OAuth security metadata, and non-gRPC transport filtering. (#499)
Add skill models and skills sync API for managing MCP server skill definitions. (#501)
Add Registry-side OpenTelemetry tracing routed to Langfuse Cloud through OTel Collector with structured operation spans for MCP tool execution and A2A agent execution. (#505)
Add Skill model fields, RegistryResourceType.SKILL, and skill CRUD endpoints with dual-auth and skill-specific scopes. (#510)
Add scope ceiling per client category to ensure MCP proxy clients and other DCR clients cannot access unauthorized A2A scopes. (#512)
Improve OAuth diagnostics and Redis configuration hygiene. (#516)
Display granted scopes on OAuth consent page to show users what permissions they are granting. (#522)
Harden workflow run monitoring with adaptive polling, dirty detection, and live node status updates. (#506)
Display scopes on downstream consent page to provide transparency during OAuth authorization flows. (#524)