proactive-task-engine

name	proactive-task-engine
slug	proactive-task-engine
description	Proactive task discovery engine. Periodically scans task boards (OKR, GitHub Issues, Notion, Linear) for items the lobster can pick up, generates structured approval requests with confidence scoring. Turns lobsters from reactive responders into self-driving agents.

Proactive Task Engine — Self-Driving Lobster

Scan → Discover → Score → Propose → Execute (on approval)

Instead of waiting for assignments, lobsters proactively find work they can do.

When to Trigger

• Cron nudge: Controller sends a daily [proactive-scan] message via proactive-cron.sh
• Heartbeat idle: During heartbeat, if no active tasks and idle for >2 hours
• Manual: Owner says "scan for tasks" / "find something to do"

Core Loop

┌──────────────┐     ┌──────────────┐     ┌──────────────┐
│  Scan Boards │ ──▶ │ Filter/Score │ ──▶ │   Propose    │
└──────────────┘     └──────────────┘     └──────────────┘
                                                 │
                                          Owner approves?
                                           │          │
                                          yes         no
                                           │          │
                                     ┌─────▼──┐  ┌───▼───┐
                                     │Execute  │  │ Skip  │
                                     └────────┘  └───────┘

Phase 1: Scan

Connect to configured task boards and pull open/unassigned items.

Supported Backends

Backend	How to Query	Configuration
Notion	Notion API — filter by Status = "Open", Assignee = empty	config.notion.databaseId
GitHub Issues	GitHub API — filter by is:open no:assignee	config.github.repo
Linear	Linear API — filter by unassigned, open state	config.linear.teamId
Local file	Read tasks.json or TODO.md in workspace	config.local.path

Scan Query

Fetch all items where:
  - Status: Open / Backlog / To Do
  - Assignee: unassigned OR assigned to me
  - Priority: any (scoring handles prioritization)
  - Updated: within last 30 days (skip stale items)

Phase 2: Filter & Score

For each discovered item, compute a fit score (0-10):

Scoring Criteria

Factor	Weight	Description
Capability match	40%	Does the task fall within my skill domain?
Priority	25%	P0/P1 score higher than P3
Effort estimate	20%	Can I complete this within 1 work session?
Dependencies	15%	Are all prerequisites met?

Capability Matching

Read your own MEMORY.md or capability profile to determine your domains:

• Code: languages, frameworks, modules you've worked on
• Research: topics you've investigated before
• Ops: infrastructure, deployment, monitoring
• Design: UI/UX, product specs

Score Thresholds

Score	Action
8-10	High confidence — include in proposal
5-7	Medium — include with caveats
0-4	Low — skip unless nothing else available

Phase 3: Propose

Generate a structured approval request for your owner (via DM or thread):

🔍 **Proactive Task Proposal**

I scanned [Board Name] and found items I can help with:

**High Confidence (8+):**
1. 🟢 [Task Title] (Score: 9/10)
   - Why: Matches my [domain] expertise, no dependencies, ~2h effort
   - Plan: [1-2 sentence approach]

2. 🟢 [Task Title] (Score: 8/10)
   - Why: [reasoning]
   - Plan: [approach]

**Medium Confidence (5-7):**
3. 🟡 [Task Title] (Score: 6/10)
   - Why: Related to my domain but involves [unfamiliar area]
   - Plan: [approach with caveats]

Shall I proceed with any of these? Reply with numbers (e.g., "do 1 and 2")
or "skip" to pass.

Proposal Rules

• Maximum 5 items per proposal (don't overwhelm)
• Always include your reasoning (why you think you can do it)
• Always include effort estimate
• If nothing scores above 5: report "scanned, nothing actionable" — don't force it

Phase 4: Execute

On owner approval:

1. Update task board: assign to self, set status to "In Progress"
2. Execute the task using your skills and tools
3. Report progress in the relevant thread/channel
4. On completion: update board status to "Done", notify owner

Configuration

Store in your workspace as proactive-config.json:

{
  "enabled": true,
  "scanInterval": "daily",
  "backends": [
    {
      "type": "notion",
      "databaseId": "YOUR_DATABASE_ID",
      "filter": {
        "status": ["Open", "Backlog"],
        "assignee": ["unassigned", "self"]
      }
    }
  ],
  "capabilities": ["backend", "api-design", "data-analysis"],
  "maxProposals": 5,
  "autoExecuteThreshold": null
}

Auto-Execute Mode (optional)

If owner sets autoExecuteThreshold (e.g., 9), tasks scoring at or above that threshold are executed without waiting for approval. Use with caution.

File Structure

skills/proactive-task-engine/
├── SKILL.md              # This file
└── references/
    └── scoring-guide.md  # Detailed scoring rubric and examples

Dependencies

• Task board API access (Notion token, GitHub token, etc.)
• Read access to own capability profile / MEMORY.md
• Message tool for sending proposals to owner