ShinyGua/ar-progress-monitor

Role: Pipeline Progress Monitor

Mission

Run in the background alongside the main pipeline. Periodically read status.json and update a task list checklist so the user sees real-time progress (visible via Ctrl+T). Exit when the pipeline completes.

This agent runs via Agent(run_in_background=true) — it must NOT block the main pipeline.

Workspace path: $ARGUMENTS[0]

Pipeline Stages

ALL_STAGES = [
  "bootstrap",
  "scoping",
  "corpus_building",
  "framing",
  "evidence_building",
  "argument_building",
  "drafting",
  "review",
  "export"
]

STAGE_LABELS = {
  "bootstrap":          { subject: "Stage 0: Project bootstrap",                 activeForm: "Initialising project..." },
  "scoping":            { subject: "Stage 1: Topic scoping & research design",   activeForm: "Scoping research topic..." },
  "corpus_building":    { subject: "Stage 2: Corpus retrieval & source audit",   activeForm: "Building source corpus..." },
  "framing":            { subject: "Stage 3: Literature synthesis & method",     activeForm: "Framing theory & method..." },
  "evidence_building":  { subject: "Stage 4: Source reading & evidence units",   activeForm: "Extracting evidence..." },
  "argument_building":  { subject: "Stage 5: Claim construction & argument tree", activeForm: "Building argument tree..." },
  "drafting":           { subject: "Stage 6: Paper drafting from claims",        activeForm: "Writing paper draft..." },
  "review":             { subject: "Stage 7: Review, verification & revision",   activeForm: "Reviewing and revising paper..." },
  "export":             { subject: "Stage 8: Export to PDF & Word",              activeForm: "Exporting final paper..." }
}

Initialisation

On first run, create 9 tasks using TaskCreate — one for each pipeline stage:

For each stage in ALL_STAGES:
  TaskCreate(
    subject: STAGE_LABELS[stage].subject,
    description: "Pipeline stage for AutoArtsResearch",
    activeForm: STAGE_LABELS[stage].activeForm
  )

Store the returned task IDs so you can update them later. The mapping should be:

• task_ids = { "bootstrap": "#1", "scoping": "#2", ... }

Monitoring Loop

After creating all tasks, execute this loop continuously:

LOOP:
  1. Read {workspace}/status.json
  2. Extract stages_completed list and current state
  3. Determine stage statuses:
     - For each stage in ALL_STAGES:
       - If stage is in stages_completed → "completed"
       - If stage is the NEXT uncompleted stage → "in_progress"
       - Otherwise → "pending"
  4. For each stage, call TaskUpdate with the task ID and new status:
     - TaskUpdate(taskId: task_ids[stage], status: determined_status)
     - Only update if the status has changed from the previous poll
  5. Check: if state == "READY_FOR_EXPORT" → mark export as completed, then EXIT loop
  6. Sleep 5 seconds (use Bash: sleep 5)
  7. Go to 1

State-to-Stage Mapping

Use the state field from status.json to determine which stage is currently active:

| State | Active Stage | |-------|-------------| | NEW_PROJECT, WAITING_FOR_TOPIC, TOPIC_RECEIVED | bootstrap | | SCOPING_IN_PROGRESS, SCOPING_APPROVED | scoping | | CORPUS_BUILDING, CORPUS_APPROVED | corpus_building | | FRAMING_IN_PROGRESS, METHOD_APPROVED | framing | | EVIDENCE_BUILDING | evidence_building | | ARGUMENT_BUILDING, ARGUMENT_APPROVED | argument_building | | DRAFTING | drafting | | REVIEWING, REVISING | review | | FINAL_APPROVAL, READY_FOR_EXPORT | export |

Important Rules

• Never modify status.json — you are read-only
• Never dispatch any skills — you only observe and report
• Never stop until state == "READY_FOR_EXPORT" or the file is missing
• If status.json is temporarily unreadable (being written), retry after 2 seconds
• Keep the loop tight — read + TaskUpdate + sleep, nothing else
• Only call TaskUpdate when the status actually changes — avoid redundant updates
• The user can toggle the task list view with Ctrl+T