rp1-run/task

Task Queue Management

Interact with the rp1 persistent task queue. Tasks are stored in ~/.rp1/rp1.db and processed in FIFO order (oldest first). The task system is a queue and state tracker -- execution is the responsibility of the agent or harness hook that picks up the task.

Task Lifecycle

Tasks move through these states:

pending -> in_progress -> completed
                       -> failed
pending -> cancelled
in_progress -> cancelled

• New tasks start as pending.
• pickup atomically transitions the oldest pending task to in_progress.
• Only in_progress tasks can be completed or failed.
• Only pending and in_progress tasks can be cancelled.

Operations

Create a Task

Queue a new task for later execution.

rp1 agent-tools task create \
  --type <type> \
  --description <text> \
  [--payload <json>] \
  [--project <path>]

| Option | Required | Description | |--------|----------|-------------| | --type | Yes | Free-form task type (e.g., check-annotations, archive-feature, remediate-audit) | | --description | Yes | Human-readable description of the work | | --payload | No | JSON string with type-specific data | | --project | No | Absolute project path for scoping (omit for global tasks) |

Example:

rp1 agent-tools task create \
  --type "check-annotations" \
  --description "Review open annotations from last audit" \
  --project /Users/dev/myapp

Example with payload:

rp1 agent-tools task create \
  --type "remediate-audit-findings" \
  --description "Fix security issues found in auth module" \
  --payload '{"files":["src/auth.ts","src/session.ts"],"severity":"high"}'

List Tasks

View tasks filtered by status and/or project.

rp1 agent-tools task list \
  [--status <status>] \
  [--project <path>] \
  [--limit <n>]

| Option | Required | Description | |--------|----------|-------------| | --status | No | Filter by status: pending, in_progress, completed, failed, cancelled | | --project | No | Filter by absolute project path | | --limit | No | Maximum number of tasks to return |

Example -- list pending tasks for a project:

rp1 agent-tools task list --status pending --project /Users/dev/myapp

Results are always ordered by creation time (oldest first).

Pick Up Next Task

Atomically claim the oldest pending task for execution.

rp1 agent-tools task pickup [--project <path>]

| Option | Required | Description | |--------|----------|-------------| | --project | No | Only pick up tasks scoped to this project path |

The returned task is automatically transitioned to in_progress. If no pending tasks exist, the response data is null (not an error).

Example:

rp1 agent-tools task pickup --project /Users/dev/myapp

Complete a Task

Mark an in-progress task as successfully completed.

rp1 agent-tools task complete --id <id> [--result <text>]

| Option | Required | Description | |--------|----------|-------------| | --id | Yes | Task ID (positive integer) | | --result | No | Summary of what was accomplished |

Example:

rp1 agent-tools task complete --id 42 --result "Reviewed 3 annotations, all resolved"

Fail a Task

Mark an in-progress task as failed.

rp1 agent-tools task fail --id <id> [--result <text>]

| Option | Required | Description | |--------|----------|-------------| | --id | Yes | Task ID (positive integer) | | --result | No | Error description explaining the failure |

Example:

rp1 agent-tools task fail --id 42 --result "Could not access auth module: file not found"

Cancel a Task

Cancel a pending or in-progress task.

rp1 agent-tools task cancel --id <id>

| Option | Required | Description | |--------|----------|-------------| | --id | Yes | Task ID (positive integer) |

Only tasks in pending or in_progress state can be cancelled. Completed and failed tasks cannot be cancelled.

Example:

rp1 agent-tools task cancel --id 42

Get a Task

Retrieve a single task by ID.

rp1 agent-tools task get --id <id>

| Option | Required | Description | |--------|----------|-------------| | --id | Yes | Task ID (positive integer) |

Example:

rp1 agent-tools task get --id 42

Response Format

All commands return a ToolResult JSON envelope:

{
  "success": true,
  "tool": "task",
  "data": {
    "id": 1,
    "type": "check-annotations",
    "description": "Review open annotations from last audit",
    "status": "pending",
    "payload": null,
    "projectPath": "/Users/dev/myapp",
    "result": null,
    "createdAt": "2026-03-15T10:30:00.000Z",
    "updatedAt": "2026-03-15T10:30:00.000Z"
  }
}

On error:

{
  "success": false,
  "tool": "task",
  "data": null,
  "errors": [{ "message": "Task 99 not found" }]
}

For list, data is an array of task records. For pickup with no pending tasks, data is null with success: true.

Task Processing Loop

Recommended pattern for harness hooks that process queued tasks:

# 1. Pick up the next pending task
RESULT=$(rp1 agent-tools task pickup --project "$PROJECT_PATH")

# 2. Check if a task was returned
TASK_ID=$(echo "$RESULT" | jq -r '.data.id // empty')
if [ -z "$TASK_ID" ]; then
  echo "No pending tasks"
  exit 0
fi

# 3. Read task details
TASK_TYPE=$(echo "$RESULT" | jq -r '.data.type')
TASK_DESC=$(echo "$RESULT" | jq -r '.data.description')
TASK_PAYLOAD=$(echo "$RESULT" | jq -r '.data.payload // empty')

# 4. Execute based on task type and report outcome
#    (agent determines action from type + description + payload)
#    On success:
rp1 agent-tools task complete --id "$TASK_ID" --result "Done: processed successfully"
#    On failure:
rp1 agent-tools task fail --id "$TASK_ID" --result "Error: <description>"

Validation Rules

• --type and --description must be non-empty strings
• --project must be an absolute path (starts with /) when provided
• --payload must be valid JSON when provided
• --status must be one of: pending, in_progress, completed, failed, cancelled
• --id must be a positive integer