- name
- task-extractor
- description
- Extract, track, and verify completion of multiple tasks from a single user message. Use when a message contains 3+ actionable items or mixes multiple unrelated asks. Prevents the common failure mode where agents address 2 of 6 requests and silently drop the rest.
Task Extractor
When to Trigger
A user message contains 3+ actionable items, mixes asks across different domains, or has a history of dropped tasks.
Process
- Extract every task from the message — err on the side of capturing more.
- Number each task explicitly.
- Track in a hot index:
| # | Task | Status | Evidence | - Execute each in order (or delegate to appropriate skill).
- Reconcile after completion — compare extracted list against actual completions.
Example
User: "Fix the deploy bug, update the landing page copy,
check if Stripe webhook is working, and email the client"
Extracted:
1. Fix deploy bug → coder
2. Update landing page copy → coder
3. Check Stripe webhook → verifier
4. Email client → GATE (needs approval)Rules
- Missing a task is worse than capturing an extra one.
- Mark each explicitly: ✅ done | 🔄 in progress | ❌ blocked | ⏭️ deferred.
- At end of run, list any DROPPED tasks — these are the most important to surface.
- If a task needs human approval, mark as GATE and don't execute silently.