Problem / Motivation
_load_agent_cards() in workbench/app.py reads agent metadata from the AGENT_DIRECTORY env var (a JSON blob). This is a static snapshot that must be manually kept in sync with what's actually deployed. kagent already maintains a live agent registry from Kubernetes CRDs and exposes it via API.
Proposed Solution
Replace the env var lookup with a call to kagent's agent listing API. The workbench becomes a consumer of kagent's registry rather than maintaining its own.
Changes:
workbench/app.py:_load_agent_cards() — call kagent API instead of parsing env var
workbench/app.py:agent_directory() — pass through kagent response, optionally enrich with studio-specific metadata
agents/assistant/delegate.py:_resolve_agent_url() — resolve via kagent instead of workbench directory
Acceptance Criteria
Related
Problem / Motivation
_load_agent_cards()inworkbench/app.pyreads agent metadata from theAGENT_DIRECTORYenv var (a JSON blob). This is a static snapshot that must be manually kept in sync with what's actually deployed. kagent already maintains a live agent registry from Kubernetes CRDs and exposes it via API.Proposed Solution
Replace the env var lookup with a call to kagent's agent listing API. The workbench becomes a consumer of kagent's registry rather than maintaining its own.
Changes:
workbench/app.py:_load_agent_cards()— call kagent API instead of parsing env varworkbench/app.py:agent_directory()— pass through kagent response, optionally enrich with studio-specific metadataagents/assistant/delegate.py:_resolve_agent_url()— resolve via kagent instead of workbench directoryAcceptance Criteria
AGENT_DIRECTORYenv var removed or made fallback-onlyRelated