Skip to content
Tandem Engine

Agent Team API

Endpoints

GET /agent-team/templates

Returns loaded agent templates.

GET /agent-team/instances

Query params:

  • missionID (optional)
  • parentInstanceID (optional)
  • status (optional: queued|running|completed|failed|cancelled)

GET /agent-team/missions

Returns mission-level rollups across instances:

  • instanceCount
  • runningCount
  • completedCount
  • failedCount
  • cancelledCount
  • queuedCount
  • tokenUsedTotal
  • toolCallsUsedTotal
  • stepsUsedTotal
  • costUsedUsdTotal

GET /agent-team/approvals

Returns:

  • spawnApprovals: queued spawn approvals (request-only edges / approval-required paths)
  • toolApprovals: pending permission requests scoped to agent-team sessions

POST /agent-team/spawn

Request body:

{
  "missionID": "m1",
  "parentInstanceID": "ins_parent",
  "templateID": "worker-default",
  "role": "worker",
  "source": "ui_action",
  "justification": "Implement focused task",
  "budget_override": {
    "max_tokens": 8000,
    "max_steps": 4,
    "max_tool_calls": 10,
    "max_duration_ms": 240000,
    "max_cost_usd": 1.25
  }
}

Success response:

{
  "ok": true,
  "missionID": "m1",
  "instanceID": "ins_...",
  "sessionID": "ses_...",
  "runID": null,
  "status": "running",
  "skillHash": "sha256:..."
}

Error response:

{
  "ok": false,
  "code": "spawn_policy_missing",
  "error": "spawn policy file missing",
  "requiresUserApproval": false
}

If policy requires user approval, response shape is still 200 with:

{
  "ok": false,
  "code": "spawn_requires_approval",
  "requiresUserApproval": true
}

Tool-driven spawn (spawn_agent)

spawn_agent uses the same server-side policy/runtime gate as POST /agent-team/spawn.

Invoke it through a session prompt:

/tool spawn_agent {"missionID":"m1","role":"worker","templateID":"worker-default","source":"tool_call","justification":"parallelize implementation"}

Tool response metadata shape:

{
  "ok": true,
  "missionID": "m1",
  "instanceID": "ins_...",
  "sessionID": "ses_...",
  "runID": null,
  "status": "running",
  "skillHash": "sha256:..."
}

POST /agent-team/instance/{id}/cancel

Request body:

{
  "reason": "manual stop"
}

POST /agent-team/approvals/spawn/{id}/approve

Request body:

{
  "reason": "approved from command center"
}

Response:

{
  "ok": true,
  "approvalID": "appr_...",
  "decision": "approved",
  "instanceID": "ins_...",
  "missionID": "m1",
  "status": "running"
}

POST /agent-team/approvals/spawn/{id}/deny

Request body:

{
  "reason": "not aligned with mission scope"
}

Response:

{
  "ok": true,
  "approvalID": "appr_...",
  "decision": "denied",
  "reason": "not aligned with mission scope"
}

Response:

{
  "ok": true,
  "instanceID": "ins_...",
  "sessionID": "ses_...",
  "status": "cancelled"
}

POST /agent-team/mission/{id}/cancel

Request body:

{
  "reason": "stop all mission agents"
}

Response:

{
  "ok": true,
  "missionID": "m1",
  "cancelledInstances": 3
}

POST /mission/{id}/event (orchestrator bridge fields)

When mission runtime bridge is active, mission event responses include additive fields:

  • orchestratorSpawns: spawn results triggered by mission_started
  • orchestratorCancellations: cancellation summary triggered by mission_canceled