List agent evaluation runs

Outcome

Reads up to fifty sandbox evaluation runs with their per-case results. It does not start an evaluation or invoke a model.

Before you start

• Verify the intended environment, authenticated identity and organization before discovery.
• Discover the published agents:read application permission and inspect the credential-scoped action contract.
• Discover and bind the granted canonical Agent workspace separately from the action input.
• Fixture identifiers and simulated responses confer no live access. Replace identifiers through granted resource discovery.

Discover the target

• topolo whoami --json
• topolo workspaces list topolo-agent --json
• topolo actions get app_topolo_agent.agents.evaluation_runs.list --detail --json
• Read app_topolo_agent.agents.list in this same authorized context and select the returned identifier. Do not derive ids from display names.
• Inspect app_topolo_agent.agents.evaluation_suite.get in the same credential and resource context before reading it.

Request

GET /api/agents/{id}/evaluation-runs

Input fields

Worked examples

Worked response 1 · HTTP 200

Actual exported read handlers and replayed SQLite migrations. Auth directory, service-token, R2 object storage and Durable Object storage boundaries are simulated; session response serialization executes the actual AgentSession.onRequest method. Egress is sealed. No deployed/provider outcome or live access is demonstrated.

Input · payload.json:

{
  "id": "agent_docs_fixture"
}

CLI:

topolo actions validate app_topolo_agent.agents.evaluation_runs.list --data @payload.json --json
topolo actions plan app_topolo_agent.agents.evaluation_runs.list --resource-type workspace --resource-id "$AGENT_WORKSPACE_ID" --data @payload.json --json
topolo actions call app_topolo_agent.agents.evaluation_runs.list --resource-type workspace --resource-id "$AGENT_WORKSPACE_ID" --data @payload.json --json

MCP tool and arguments:

{
  "tool": "topolo_read_action",
  "arguments": {
    "actionId": "app_topolo_agent.agents.evaluation_runs.list",
    "input": {
      "id": "agent_docs_fixture"
    },
    "resourceType": "workspace",
    "resourceId": "<identifier returned by granted resource discovery>"
  }
}

Captured local response · HTTP 200:

{
  "success": true,
  "agentId": "agent_docs_fixture",
  "sandboxRuns": []
}

successful: HTTP 200; seeded local fixture identifiers confer no live access.

Worked response 2 · HTTP 403

Actual exported read handlers and replayed SQLite migrations. Auth directory, service-token, R2 object storage and Durable Object storage boundaries are simulated; session response serialization executes the actual AgentSession.onRequest method. Egress is sealed. No deployed/provider outcome or live access is demonstrated.

Input · payload.json:

{
  "id": "agent_docs_fixture"
}

CLI:

topolo actions validate app_topolo_agent.agents.evaluation_runs.list --data @payload.json --json
topolo actions plan app_topolo_agent.agents.evaluation_runs.list --resource-type workspace --resource-id "$AGENT_WORKSPACE_ID" --data @payload.json --json

MCP tool and arguments:

{
  "tool": "topolo_prepare_action",
  "arguments": {
    "actionId": "app_topolo_agent.agents.evaluation_runs.list",
    "input": {
      "id": "agent_docs_fixture"
    },
    "resourceType": "workspace",
    "resourceId": "<identifier returned by granted resource discovery>"
  }
}

Captured local response · HTTP 403:

{
  "success": false,
  "error": "Insufficient permissions"
}

permission denied: HTTP 403; SDK code permission_denied; seeded local fixture identifiers confer no live access.

Worked response 3 · HTTP 404

Actual exported read handlers and replayed SQLite migrations. Auth directory, service-token, R2 object storage and Durable Object storage boundaries are simulated; session response serialization executes the actual AgentSession.onRequest method. Egress is sealed. No deployed/provider outcome or live access is demonstrated.

Input · payload.json:

{
  "id": "agent_docs_fixture"
}

CLI:

topolo actions validate app_topolo_agent.agents.evaluation_runs.list --data @payload.json --json
topolo actions plan app_topolo_agent.agents.evaluation_runs.list --resource-type workspace --resource-id "$AGENT_WORKSPACE_ID" --data @payload.json --json

MCP tool and arguments:

{
  "tool": "topolo_prepare_action",
  "arguments": {
    "actionId": "app_topolo_agent.agents.evaluation_runs.list",
    "input": {
      "id": "agent_docs_fixture"
    },
    "resourceType": "workspace",
    "resourceId": "<identifier returned by granted resource discovery>"
  }
}

Captured local response · HTTP 404:

{
  "success": false,
  "error": "Agent not found."
}

private resource hidden: HTTP 404; SDK code http_error; seeded local fixture identifiers confer no live access.

Verify the result

• Confirm agentId and inspect agentVersionId, status, requiredCheckCount, passedRequiredCheckCount and each case result. An empty sandboxRuns array means no recorded runs were returned.

Errors and recovery

• auth_error (HTTP 401): For an unstructured HTTP 401 response, authenticate and verify the intended organization and workspace before retrying.
• permission_denied (HTTP 403): For an unstructured HTTP 403 response, use the published application permission and required resource-level access. Do not retry with broader context.
• http_error (HTTP 404): For an unstructured HTTP 404 response, refresh the corresponding resource listing in the same context. Private resources can be deliberately hidden from this caller.

Authentication, input validation or resource middleware can refuse a request before these handlers. Preserve the actual refusal and request id.

Recovery boundary

This read does not change service state. Re-discover the target in the same authorized organization and workspace after a missing or hidden-resource refusal; do not guess identifiers or widen context.

Related actions

• Get agent evaluation suite: Read the editable draft suite and immutable current-version evaluation suite.

Evidence

Actual exported read handlers and replayed SQLite migrations. Auth directory, service-token, R2 object storage and Durable Object storage boundaries are simulated; session response serialization executes the actual AgentSession.onRequest method. Egress is sealed. No deployed/provider outcome or live access is demonstrated.