jarvis-tasks

Jarvis Task Manager

Manage and execute tasks from the Jarvis task registry.

Context

• Workspace root: resolved from CLAUDE.md canonical root. In Cowork sessions, use mounted workspace (visible in session environment).
• Task registry: Jarvis/tasks.json
• Architecture doc: Jarvis/architecture.md
• Task output directory: Jarvis/analysis/ (analysis tasks), Jarvis/audit/ (maintenance tasks)

Boot

Read Jarvis/tasks.json to load all task definitions.

Task Menu

Present this menu to Chadron:

JARVIS TASK MANAGER
─────────────────────────────────
[1] List All Tasks       View tasks by category with status
[2] Run a Task           Execute a specific task by name or ID
[3] Run Category         Execute all tasks in a category
[4] Task History         View recent task execution results
[5] Quick Run            Run a recommended task based on current state
─────────────────────────────────

Option 1: List All Tasks

Read Jarvis/tasks.json and display tasks grouped by category:

DATA OPERATIONS
─────────────────────────────────
  refresh-itsm-core      Refresh ITSM Core Data         [10 min]  Last: never
  refresh-full           Full ServiceNow Refresh         [30 min]  Last: never  ⚠ Approval
  refresh-catalog        Refresh Catalog Data            [8 min]   Last: never

ANALYSIS & REPORTING
─────────────────────────────────
  analyze-incident-patterns   Analyze Incident Patterns  [15 min]  Last: never
  analyze-resolution-times    Analyze Resolution Times   [12 min]  Last: never
  analyze-automation-scan     Automation Opportunity Scan [20 min]  Last: never
  generate-weekly-report      Weekly Status Report       [10 min]  Last: never  📅 Mon
  generate-weekly-digest      Weekly Digest              [8 min]   Last: never  📅 Fri

[... continue for all categories]

Option 2: Run a Task

When Chadron specifies a task:

1. Load the task definition from tasks.json
2. Show task details: name, description, estimated time, inputs, outputs
3. If task has configurable inputs, present them with defaults
4. If approval_required: true, explicitly confirm before proceeding
5. Execute the task by following the agent_prompt exactly as written
6. Update last_run and last_status in tasks.json when complete

Execution Protocol

When executing a task:

• Follow the agent_prompt field as your primary instructions
• Use only the tools listed in tools_required
• Save outputs to the paths specified in outputs.files
• Update brain files listed in outputs.brain_updates
• Report progress as you go (which step you're on, what you found)

After Execution

TASK COMPLETE: [task name]
─────────────────────────────────
Duration:    [actual time]
Status:      ✓ Success / ✗ Failed
Output:      [list of files created/updated]
Brain Updates: [list of brain files modified]
─────────────────────────────────

Update tasks.json with:

• last_run: current ISO timestamp
• last_status: "success" or "failed"

Option 3: Run Category

List categories and run all tasks in the selected category sequentially:

• Skip tasks that have approval_required: true (list them as skipped)
• Report progress after each task
• Stop if any task fails

Option 4: Task History

Read tasks.json and display all tasks with non-null last_run:

RECENT TASK RUNS
─────────────────────────────────
[date]  analyze-incident-patterns   ✓ Success   12 min
[date]  check-data-freshness       ✓ Success    1 min
[date]  refresh-itsm-core          ✗ Failed     3 min
─────────────────────────────────

Option 5: Quick Run

Assess current state and recommend the most valuable task to run next:

1. Read status.json — check data freshness
2. Read recommendations.md — check top priorities
3. Read tasks.json — check what hasn't been run recently

Recommend based on:

• Data freshness > 7 days → suggest data refresh task
• Weekly maintenance not run this week → suggest maintenance
• KB articles at 5% → suggest KB drafting task
• Catalog design not started → suggest catalog design task

Present the recommendation and ask to proceed.

Adding New Tasks

If Chadron describes a new task to add:

1. Generate a task definition matching the tasks.json schema
2. Assign appropriate category, priority, and tools
3. Write the agent_prompt with full context and instructions
4. Add to tasks.json
5. Confirm the task is available

Rules

• Always read tasks.json fresh before any operation (it may have been updated)
• Never modify agent_prompt fields of existing tasks without explicit request
• Always update last_run and last_status after execution
• Respect approval_required — never skip confirmation for flagged tasks
• Report all outputs clearly so Chadron can review