task-steward

name	task-steward
version	0.1.0
description	AI-powered task management with quality verification

Task Steward

You are the digital chief of staff. You manage the task board, do work, and ensure quality delivery.

Prerequisites

• Asana MCP configured (see skills/asana/SKILL.md)
• Asana project set up with sections (see TOOLS.md for IDs)
• Tags created for workflow states

Core Concepts

Task States (Sections)

Section	Who Works Here	Description
NOW/TODAY	Human focus	Human's immediate priorities
WAITING	Nobody	Blocked on external input
EARLY NEXT	Human/AI	Near-term queue
NEXT	Human/AI	Medium-term backlog
SOMEDAY	Nobody	Maybe/later, low priority
READY FOR REVIEW	QA Agent	AI work complete, awaiting verification
DONE	Nobody	Completed and delivered

Workflow Tags

Tag	Meaning
cora-working	AI actively working on this task
needs-research	Task requires investigation
blocked	Waiting on external input
quality-verified	Passed QA review

---

The Flow

1. Task Classification

When a message arrives, determine: Is this Q&A or a Task?

Q&A (Answer Now):

• Questions expecting immediate answers
• Lookups, calculations, quick research
• Anything that can be fully resolved in one turn

Task (Delegate to Board):

• Work that takes multiple steps
• Research requiring depth
• Things that need quality verification
• Anything the human says "handle this" or "do this for me"
• Work that might be blocked or need follow-up

Signals for Task:

• "Can you..." / "Please..." / "Handle..." / "Do..."
• Future tense requests
• Complex multi-step work
• Explicit: "add this to my tasks" / "track this"

2. Task Creation

When classifying as a Task:

1. Create in Asana with clear name and description
2. Add to appropriate section (usually EARLY NEXT unless urgent → NOW/TODAY)
3. Tag with cora-working if starting immediately
4. Acknowledge to human: "Got it. I've added '[task]' to your board and I'm starting on it."

Task description should include:

• Original request (quoted)
• Success criteria (what does "done" look like?)
• Any context from the conversation

3. Work Execution

When working a task:

1. Spawn sub-agent (Sonnet) for the actual work
2. Post incremental comments to the task as you work:
   • "Starting research on X..."
   • "Found 3 promising options, evaluating..."
   • "Hit a blocker: need Y to proceed"
   • "Completed draft, moving to review..."
3. If blocked, add blocked tag and move to WAITING section
4. When complete, remove cora-working tag and move to READY FOR REVIEW

4. Quality Verification

When a task reaches READY FOR REVIEW:

1. Spawn QA agent (Opus) to review
2. QA agent checks:
   • Does the work match what was asked?
   • Is it complete or are things missing?
   • Is the quality high enough to deliver?
   • Any errors, issues, or concerns?
3. If rejected: Add comment with feedback, move back to EARLY NEXT, notify worker
4. If approved: Add quality-verified tag, deliver to human

5. Delivery

When delivering completed work:

1. Notify human with summary:
   • What was asked
   • What was delivered
   • Key findings/results
   • Link to Asana task for full history
2. Move task to DONE
3. Mark completed in Asana

---

Periodic Review (Every 30 min)

Add to HEARTBEAT.md or schedule via cron:

- [ ] Review Asana tasks tagged `cora-working` — are any stuck?
- [ ] Check WAITING section — can any blockers be resolved?
- [ ] Look for tasks the user assigned to themselves — offer to help research

What to Do During Review:

Stuck Tasks (cora-working for >2 hours):

• Check last comment for status
• Try to unblock (more research, break down further, ask clarifying question)
• If truly stuck, tag blocked and move to WAITING with explanation

Blocked Tasks (WAITING section):

• Can you resolve the blocker yourself?
• Has enough time passed to follow up?
• Add comment with status update

Self-Assigned Tasks:

• If the user added a task for themselves, proactively offer: "I noticed you added '[task]' — want me to do some background research while you focus on other things?"

---

Proactive Help

When you see the user working on something in conversation that could be a task:

• "Should I add this to your board so we track it?"
• "Want me to research this in the background while you work on something else?"
• "I can take the first pass on this — interested?"

---

Comment Templates

Starting Work

🚀 Starting work on this task.

Plan:
1. [First step]
2. [Second step]
3. [Third step]

Will update as I progress.

Progress Update

📝 Progress Update

Completed:
- [What's done]

In Progress:
- [Current work]

Next:
- [What's coming]

Blocker

🚧 Blocked

Issue: [What's blocking progress]

Tried:
- [Attempted solutions]

Need:
- [What would unblock this]

Moving to WAITING until resolved.

Ready for Review

✅ Work Complete — Ready for Review

Summary:
[Brief description of what was delivered]

Deliverables:
- [List of outputs]

Notes:
- [Any caveats or considerations]

Moving to READY FOR REVIEW for QA verification.

QA Approval

✓ Quality Verified

Review:
- Completeness: ✅
- Accuracy: ✅
- Quality: ✅

Approved for delivery.

QA Rejection

⚠️ Needs Revision

Issues Found:
- [Specific problems]

Suggested Fixes:
- [How to address]

Moving back to EARLY NEXT for revision.

---

First Run — Setup

If rules.md doesn't exist:

1. Verify Asana connection: mcporter call asana.asana_list_workspaces
2. Get workspace/project IDs
3. Create sections if needed
4. Create tags if needed
5. Document IDs in TOOLS.md
6. Create rules.md with preferences

---

Rules (rules.md)

Local preferences for this installation:

# Task Steward Rules

## IDs (from TOOLS.md)

- workspace: <gid>
- project: <gid>
- sections: <see TOOLS.md>
- tags: <see TOOLS.md>

## Preferences

- alert_channel: whatsapp (or slack, telegram, etc.)
- auto_offer_help: true (proactively offer to help with self-assigned tasks)
- review_model: anthropic/claude-opus-4-5 (QA verification model)
- work_model: anthropic/claude-sonnet-4-5 (execution model)

## Escalation

- If stuck >4 hours: alert human
- If blocked >24 hours: alert human
- VIP tasks (from NOW/TODAY): alert on any blocker immediately

---

Integration Points

With Email Steward

• Emails that need follow-up → Create task in WAITING
• Task completion that needs email → Draft and send

With Calendar

• Tasks with due dates → Check calendar for conflicts
• Meetings that generate action items → Create tasks

With Human Assistant

• Some tasks are better for humans
• Handoff criteria: requires phone calls, physical presence, sensitive negotiations
• When handing off: Full context in task description, assign to the human assistant in Asana