Rensei docs
A2A

A2A Agent Detail

AgentCard detail, trust, skills, and health.

The agent detail page gives you a complete view of a single registered external A2A agent: its skills, current health, trust level, recent invocations, and the raw AgentCard document it published.

Navigate to it from the A2A Registry:

/<org>/<project>/factory/agents/<agentCardId>

How an agent goes from registration to dispatch

Understanding how the detail page connects to the rest of the A2A subsystem helps you know which panel to look at when something goes wrong:

Every section of the detail page is a window into one step of this flow: the Skills panel reflects what the skill index knows, Trust level affects which Cedar permit policies match, Health status determines whether the router penalises or excludes this agent, and Recent invocations is a direct view of a2a_task_instances rows.

Sections

Agent identity

The top panel shows:

  • Name - from AgentCard.name
  • URL - the normalised base URL stored at registration time
  • Trust level - editable inline (see below)
  • Health status - last known status with the timestamp of the most recent probe
  • Registered / Last seen timestamps

Skills

A table listing every skill in the agent's current AgentCard.skills array:

ColumnSource
IDskill.id - also used as the MCP verb id in the MCP bridge
Nameskill.name
Descriptionskill.description (when present)
Tagsskill.tags - used as work-type hints by the skill matcher
Input schemaBadge indicating whether the skill publishes a JSON Schema (used for MCP tool input validation)

Trust level

The trust level controls how Cedar policies evaluate calls that originate from or target this agent. You can change it from the detail page without re-registering.

Trust level changes take effect immediately on the next Cedar evaluation. There is no debounce or caching window.

Valid values: external (default for new registrations) and trusted. The system level is reserved for platform-managed agents.

Cedar policy summary

A read-only summary of the Cedar policies that currently apply to this agent within the project. Click any policy to navigate to A2A Policies for editing.

Recent invocations

A table of the most recent dispatch records from a2a_task_instances for this agent. Columns:

ColumnNotes
Session IDCorrelation key used across hook-bus events and audit logs
SkillThe skill that was invoked
Statusactive / completed / failed
Started atWall-clock dispatch time
LatencyWall-clock duration for terminal rows; blank for active rows
Work typeDecoded work-type label (stripped of dispatch-kind prefix)

Click any row to navigate to the full dispatch log filtered to this agent.

Health probe history

A chronological log of health probes against this agent's discovery endpoint. Each row shows:

  • Probe timestamp
  • Latency (ms)
  • Result: healthy / degraded / unreachable
  • Error message (when applicable)

Health probes hit <agentUrl>/.well-known/agent-card.json with a 5-second timeout. A response under 2 seconds with a valid AgentCard is healthy; over 2 seconds is degraded; any non-200 or network failure is unreachable.

Raw AgentCard JSON

The raw JSON stored in the agent_cards.card column. Useful for verifying that the platform parsed and stored the card correctly, especially when debugging skill discovery issues.

{
  "name": "code-reviewer",
  "skills": [
    {
      "id": "review",
      "name": "Code Review",
      "description": "Review a PR for correctness and style.",
      "tags": ["qa"],
      "inputSchema": {
        "type": "object",
        "properties": {
          "prUrl": { "type": "string" }
        },
        "required": ["prUrl"]
      }
    }
  ]
}

Refreshing the card

Click Refresh card to re-fetch the agent's current /.well-known/agent-card.json. If the content is unchanged, only last_fetched_at is updated. If it changed, the skill index is rebuilt and the page reloads to show the new skills list.

On failure the health_status is set to unreachable and last_error is populated. The previously stored card is preserved.

On this page