Repository Issue Activity (beta)

#2860 [FEATURE] Add an access-review RPC for authorization-aware UI actions

#2718 [FEATURE] Expose DisableStandaloneSSE (or force HTTP/1.1) for RemoteMCPServer transports

#2311 [BUG] Substrate SandboxAgent workerPoolRef is not handled as a strict WorkerPool contract

#2300 [FEATURE] support China Feishu and WeCom

#2357 [FEATURE] make bash tool command timeouts configurable

#2873 [BUG] API-exposed authentication modes fail with gateway credential injection

#2872 [BUG] themeMode.tsx exports a hook beside its provider, warning on every lint run

#2871 [BUG] Flaky chat checkpoints spec: two confirmations share a Cancel label

#2849 [BUG] A newly registered MCP server is invisible — and undeletable — until the controller reaches it

#1651 [BUG] Tool calls fail silently when using mistral.ai via OpenAI-compatible provider

#2861 [FEATURE] Explore an opt-in Agent Control Standard adapter for the kagent runtime

#2862 [FEATURE] Repository-based declarative agent authoring and scaffolding

#2608 skills-init: full-clone SHA pinning always fails (checkout treats commit SHA as pathspec, not revision)

#2211 [FEATURE] meta agent that creates other kagent declarative agents

#2190 [FEATURE] Add Claude agent harness support

#2855 [FEATURE] Claude harness drops total_cost_usd and usage from the result event, so no consumer can report what a run spent

#2854 [FEATURE] A spend budget on a scheduled run, passed to the Claude harness as --max-budget-usd

#2853 [FEATURE] A declarative form of a scheduled run, so a schedule can be managed with the agent it runs

#2851 [BUG] Zero-argument tool calls serialize as "null" after session reload in OpenAI-shaped adapters

#2850 Live schedules spec asserts a detail page that deleting redirects away from

#2704 [BUG] Substrate actor and worker lists are read whole, not paged, and cannot succeed at scale

#2716 go/adk: MakeMCPAppModelResultCallback blanks non-UI tool results, same gap #2579 fixed in Python

#2549 Agent pod serves ADK-generated agent card embedding the full system prompt, diverging from the curated a2aConfig card

#2551 Task fails silently when the MCP tool backend is unreachable: no status message, no artifacts, no error log

#2760 otel.logging Helm settings never reach agent runtimes, so the documented audit-logging path emits nothing