task-state-manager

name	task-state-manager
description	Manages long-running tasks using Linear for persistence and LangSmith for observability. Multi-session aware - any chat can see and resume tasks.

Task State Manager

Manages long-running tasks with proper state persistence across chat sessions.

Problem Solved

• Tasks get lost when chats end or sessions change
• No visibility into progress of long-running work
• Can't resume interrupted tasks
• Multiple chat sessions don't share context

Solution

Uses Linear as the source of truth for task state + LangSmith for execution tracing.

Architecture

┌─────────────┐     ┌──────────────┐     ┌─────────────┐
│  Chat       │────▶│   Linear     │◀────│  Any Chat   │
│  Session 1  │     │   (State)    │     │  Session 2  │
└─────────────┘     └──────────────┘     └─────────────┘
        │                                    │
        ▼                                    ▼
┌─────────────────────────────────────────────────────┐
│              LangSmith (Tracing)                    │
│  - Every tool call logged                           │
│  - Full execution context                           │
│  - Debug failures                                   │
└─────────────────────────────────────────────────────┘

Key Features

1. Linear Issues = Task state (Backlog → In Progress → Done)
2. LangSmith Traces = Execution details (every tool call, decision, error)
3. Slack Notifications = Real-time updates on progress
4. Multi-Session = Any chat can query/resume any task

Usage

When Starting a Long Task

I'll automatically:

1. Create Linear issue in "Suelo Tasks" project
2. Start LangSmith trace
3. Send Slack notification
4. Update every 5-10 minutes

During Execution

Progress updates go to:

• Linear issue description (visible in Linear app)
• LangSmith trace (visible in smith.langchain.com)
• Slack message (if configured)

When You Ask "Did You Finish?"

I'll:

1. Query Linear for active tasks
2. Show current status + progress
3. Provide link to full trace in LangSmith
4. Offer to resume if interrupted

Setup Status

✅ Linear: Connected (Team: DEV, Project: Suelo Tasks)
✅ LangSmith: API key configured
⏳ Testing: Ready to test

Quick Test

# Create a test task
curl -X POST https://api.linear.app/graphql \
  -H "Authorization: $LINEAR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query": "mutation { issueCreate(input: { title: \"[Suelo] My Task\" description: \"Testing...\" teamId: \"29f5c661-da6c-4bfb-bd48-815a006ccaac\" projectId: \"4ad23224-d294-42ac-a9c4-1f4ac943f6e3\" }) { success issue { id identifier title url } } }"}'

Files

• task_manager.py - Core task lifecycle
• linear_client.py - Linear API wrapper
• langsmith_tracer.py - Execution tracing
• SETUP.md - Setup instructions

Projects

• Linear: "'suelo Tasks" (separate from Mercury coding tasks)
• LangSmith: "'suelo"
• Slack: #suelo

Example Workflow

You: scrape 200 instagram leads for final expense agents

Me: 
1. Creates Linear issue: "[Suelo] Scrape 200 Instagram leads"
2. Starts LangSmith trace
3. Sends Slack: "Starting task: Scrape 200 Instagram leads"
4. Works on task, updates every 10 min

[Chat ends, 2 hours pass]

You: (new chat) did you finish that instagram task?

Me:
1. Queries Linear for active tasks
2. Finds the task, sees 145/200 completed
3. Shows: "Still running - 145/200 leads found (72%)"
4. Offers: "Continue working? View trace? Show results so far?"

Benefits

✅ Never lose track - Linear is persistent
✅ Any session - Query/resume from any chat
✅ Full visibility - LangSmith shows every step
✅ Notifications - Slack keeps you updated
✅ Debugging - Traces show exactly what happened

Integration with Memory

This skill is referenced in MEMORY.md under Long-Running Task Tracking.

Before starting any task >10 minutes, I will:

1. Use this skill to create proper tracking
2. Update progress regularly
3. Mark complete when done

If you ask "did you finish that task?", I'll check Linear first.