Stations

A Station is defined by its trigger — the configuration that determines how the Job is dispatched and what the Agent receives.

Field	Type	Required	Description
id	string	yes	Unique identifier within the Assembly Line (e.g. "security")
name	string	yes	Display name (e.g. "SECURITY")
trigger	StationTrigger	yes	Dispatch configuration (see below)

---

The trigger defines whether and how a Job is dispatched when a Task arrives at this Station.

Field	Type	Default	Description
type	"dispatch_worker" | "none"	"none"	dispatch_worker creates a Job and sends it to a Runner; none is a manual or non-automated station
labels	string[]	[]	Required Runner labels for this Station (see Label Matching below)
promptTemplate	string	""	The Agent's instruction. Supports {{task.title}} and {{task.description}} interpolation
appendSystemPrompt	string	""	Additional text appended to the Agent's system prompt
idleTimeoutMinutes	number	2	Cancel the Job if the Agent is idle for this many minutes
maxTimeoutMinutes	number	60	Hard wall-clock limit for the entire Station execution
initBranch	boolean	false	If true, the Runner creates a git branch named task-{taskId} in the repository before starting the Agent
autoGit	boolean	false	If true, the Agent's session is blocked from ending if there are uncommitted changes
commitMessageTemplate	string	""	Hint injected into the Agent's context when autoGit blocks
createAssemblyLineRepo	boolean	false	Provision an Assembly Line Repository for this Task when it arrives here. Enable on the first Station that needs it.

---

{
  "id": "security",
  "name": "SECURITY",
  "trigger": {
    "type": "dispatch_worker",
    "labels": ["vibecheck", "claude-code"],
    "promptTemplate": "You are a security expert reviewing a codebase.\n\nTask: {{task.title}}\n\n{{task.description}}\n\nClone the Assembly Line Repository:\n  git clone $ASSEMBLY_LINE_REPO_URL workspace\n  cd workspace\n\nThe target repository is in ./target/\n\nPerform a thorough security audit. Write your findings to reports/security.md.\nCommit and push when done.",
    "appendSystemPrompt": "You are working as part of an automated code review pipeline. Write structured, actionable reports. Always commit and push your outputs to the Assembly Line Repository before exiting.",
    "idleTimeoutMinutes": 30,
    "maxTimeoutMinutes": 60,
    "createAssemblyLineRepo": false
  }
}

---

When the Server dispatches a Job for a Station, it builds an Agent Definition and sends it to the Runner. The Agent Definition is derived from the Station trigger plus Task-specific data.

interface AgentDefinition {
  // Routing
  labels: string[];

  // Identity
  taskId: string;
  stageId: string;   // Station ID being executed

  // Prompt (template rendered with task data)
  prompt: string;    // promptTemplate after interpolating {{task.title}}, {{task.description}}
  appendSystemPrompt?: string;

  // Workspace
  repository?: string;   // Git repo to clone (if configured at project level)
  branch?: string;
  initBranch?: boolean;
  autoGit?: boolean;
  commitMessageTemplate?: string;
  devcontainerFiles?: Record<string, string>;

  // Assembly Line Repository
  assemblyLineRepoUrl?: string;   // injected as ASSEMBLY_LINE_REPO_URL
  assemblyLineRepoToken?: string; // injected as ASSEMBLY_LINE_REPO_TOKEN (not logged)

  // Timeouts
  idleTimeoutMinutes?: number;
  maxTimeoutMinutes?: number;
}

---

Labels are how the Server routes Jobs to capable Clients. The rule is:

A Runner's registered labels must be a superset of the Station's required labels.

Station requires:  ["linux", "claude-code"]

Runner A labels:   ["linux", "claude-code"]            ✓ matches
Runner B labels:   ["linux", "claude-code", "x86_64"]  ✓ matches (superset)
Runner C labels:   ["linux"]                           ✗ no match
Runner D labels:   ["macos", "claude-code"]            ✗ no match (linux required)

When multiple Runners match, the Server dispatches to the first available idle Runner.

---

The promptTemplate field supports these template variables:

Variable	Source
{{task.title}}	TaskData.title
{{task.description}}	TaskData.description

The rendered prompt is what the Agent receives. Assembly Line designers use the description field to carry all structured input from the Task Card through to the Agent.

---

Set createAssemblyLineRepo: true on the Station where you want the repository provisioned. The Server creates it before dispatching the Job. All subsequent Stations receive assemblyLineRepoUrl and assemblyLineRepoToken in their Agent Definition automatically — no need to set createAssemblyLineRepo on every Station.

Best practice: enable it on the first Station that writes to the repository (typically INTAKE).