Inspect and recover a workspace service

workflow-workspace-recovery

The agent inspects runtime state before one bounded recovery action and reports it.

Prompt

Read task context and inspect the workspace runtime. Stop service-runner-lab exactly once, then record one progress update saying the service was stopped for recovery. Do not finish the task.

Fixture

../fixtures/companies/orientation.json

Authority

{
  "actorId": "actor-engineer",
  "capabilities": [
    "workspace:read",
    "workspace:control"
  ],
  "explicitClaims": [
    "workspace:read",
    "workspace:control"
  ],
  "taskId": "task-orientation"
}

Current results

Model	Configuration	Result	Estimated cost	Run time
claude-sonnet-5	live-acpx-0.13.1-claude-acp-0.70.0-sonnet-5	INFRA	unknown	2026-09-06T20:52:24.555Z
claude-sonnet-5	live-claude-managed-agents-2026-04-01-sonnet-5	PASS	$0.017293 estimated	2026-09-06T20:53:50.060Z
global.anthropic.claude-sonnet-4-6	live-aws-agentcore-global-anthropic-claude-sonnet-4-6	PASS	$0.062154 estimated	2026-09-06T20:52:27.156Z
gpt-5.4-mini	live-codex-pinned	INFRA	unknown	unknown
gpt-5.6-sol	live-acpx-0.13.1-codex-acp-1.6.2-control	Not run	—	—
openrouter/anthropic/claude-sonnet-5	live-opencode-1.18.17-openrouter-claude-sonnet-5	PASS	$0.013969 estimated	2026-09-06T20:55:24.790Z
openrouter/deepseek/deepseek-v4-flash-0731	live-opencode-1.18.17-openrouter-deepseek-v4-flash-0731	PASS	$0.000501 estimated	2026-09-06T20:58:37.447Z
openrouter/google/gemini-3.8-flash	live-opencode-1.18.17-openrouter-gemini-3.8-flash	PASS	$0.034436 estimated	2026-09-06T20:55:42.284Z
openrouter/openai/gpt-6-astra	live-opencode-1.18.17-openrouter-gpt-6-astra	PASS	$0.045803 estimated	2026-09-06T20:57:17.084Z
openrouter/qwen/qwen3.8-max-0902	live-opencode-1.18.17-openrouter-qwen3.8-max-0902	PASS	$0.019988 estimated	2026-09-06T20:57:25.333Z
openrouter/z-ai/glm-5.3	live-opencode-1.18.17-openrouter-glm-5.3	PASS	$0.006847 estimated	2026-09-06T20:56:12.124Z

Assertions

context

semantic_operation

{
  "id": "context",
  "kind": "semantic_operation",
  "minCalls": 1,
  "operationId": "get_task_context",
  "requireSuccess": true
}

runtime

semantic_operation

{
  "id": "runtime",
  "kind": "semantic_operation",
  "minCalls": 1,
  "operationId": "get_workspace_runtime",
  "requireSuccess": true
}

control

semantic_operation

{
  "id": "control",
  "kind": "semantic_operation",
  "maxCalls": 1,
  "minCalls": 1,
  "operationId": "control_workspace_service",
  "requireSuccess": true
}

progress

semantic_operation

{
  "id": "progress",
  "kind": "semantic_operation",
  "maxCalls": 1,
  "minCalls": 1,
  "operationId": "report_progress",
  "requireSuccess": true
}

inspect-before-control

operation_order

{
  "after": "control_workspace_service",
  "before": "get_workspace_runtime",
  "id": "inspect-before-control",
  "kind": "operation_order"
}

control-before-progress

operation_order

{
  "after": "report_progress",
  "before": "control_workspace_service",
  "id": "control-before-progress",
  "kind": "operation_order"
}

stopped

state_match

{
  "collection": "workspaceServices",
  "id": "stopped",
  "kind": "state_match",
  "recordId": "service-runner-lab",
  "where": {
    "status": "stopped"
  }
}

no-finish

semantic_operation

{
  "id": "no-finish",
  "kind": "semantic_operation",
  "maxCalls": 0,
  "operationId": "finish_task"
}

terminal-completed

terminal

{
  "id": "terminal-completed",
  "kind": "terminal",
  "status": "completed"
}

mock-only

network

{
  "id": "mock-only",
  "kind": "network",
  "realPaperclipRequests": 0
}

Full authored test definition

{
  "authority": {
    "actorId": "actor-engineer",
    "capabilities": [
      "workspace:read",
      "workspace:control"
    ],
    "explicitClaims": [
      "workspace:read",
      "workspace:control"
    ],
    "taskId": "task-orientation"
  },
  "checks": [
    {
      "id": "context",
      "kind": "semantic_operation",
      "minCalls": 1,
      "operationId": "get_task_context",
      "requireSuccess": true
    },
    {
      "id": "runtime",
      "kind": "semantic_operation",
      "minCalls": 1,
      "operationId": "get_workspace_runtime",
      "requireSuccess": true
    },
    {
      "id": "control",
      "kind": "semantic_operation",
      "maxCalls": 1,
      "minCalls": 1,
      "operationId": "control_workspace_service",
      "requireSuccess": true
    },
    {
      "id": "progress",
      "kind": "semantic_operation",
      "maxCalls": 1,
      "minCalls": 1,
      "operationId": "report_progress",
      "requireSuccess": true
    },
    {
      "after": "control_workspace_service",
      "before": "get_workspace_runtime",
      "id": "inspect-before-control",
      "kind": "operation_order"
    },
    {
      "after": "report_progress",
      "before": "control_workspace_service",
      "id": "control-before-progress",
      "kind": "operation_order"
    },
    {
      "collection": "workspaceServices",
      "id": "stopped",
      "kind": "state_match",
      "recordId": "service-runner-lab",
      "where": {
        "status": "stopped"
      }
    },
    {
      "id": "no-finish",
      "kind": "semantic_operation",
      "maxCalls": 0,
      "operationId": "finish_task"
    },
    {
      "id": "terminal-completed",
      "kind": "terminal",
      "status": "completed"
    },
    {
      "id": "mock-only",
      "kind": "network",
      "realPaperclipRequests": 0
    }
  ],
  "description": "The agent inspects runtime state before one bounded recovery action and reports it.",
  "fixture": "../fixtures/companies/orientation.json",
  "id": "workflow-workspace-recovery",
  "prompt": "Read task context and inspect the workspace runtime. Stop service-runner-lab exactly once, then record one progress update saying the service was stopped for recovery. Do not finish the task.",
  "schema": "paperclip-runner/eval-case/v1",
  "title": "Inspect and recover a workspace service"
}