Skill70 repo starsupdated 2mo ago

long-running-task-management

This skill manages long-running tasks lasting over 30 minutes or spanning multiple sessions by breaking them into checkpointed stages with resume capability. It maintains a YAML state file tracking task progress, stages, and checkpoints, enabling interruption and resumption without starting over. Use it when tasks modify many files across directories or are expected to span multiple work sessions.

View source Repository: openclaw-superpowers

Install in Claude Code

Copy

git clone --depth 1 https://github.com/ArchieIndian/openclaw-superpowers /tmp/long-running-task-management && cp -r /tmp/long-running-task-management/skills/openclaw-native/long-running-task-management ~/.claude/skills/long-running-task-management

Then start a new Claude Code session; the skill loads automatically.

Definition

SKILL.md

# Long-Running Task Management

State file: `~/.openclaw/skill-state/long-running-task-management/state.yaml`

## When to Use

- Task estimated at more than 30 minutes
- Task will span multiple sessions
- Task modifies many files across multiple directories

## Starting a Task

1. Write initial state to the state file:
   - `task_id`: short kebab-case name
   - `status: in_progress`
   - `description`: one-sentence goal
   - `stages`: ordered list with `status: pending` for each
   - `started_at`: current timestamp
2. Begin the first stage

## At Each Checkpoint

1. Complete the stage
2. Run tests/verification
3. Update state file: mark stage `status: complete`, write `checkpoint` (what's stable now), write `next_action` (first thing to do on resume), update `last_updated`
4. Commit progress to git if applicable

## Resume After Interruption

1. Read the state file
2. Check `status` and `next_action`
3. Continue from the next `pending` stage — do NOT start over

## Completion

1. Update state: `status: complete`, final `checkpoint`
2. Run full verification

## Cron Wakeup Behavior

On each 15-minute wakeup:
- Read state file
- If `status: in_progress` and `last_updated` is stale (>30 min ago): log a checkpoint update to daily memory
- If `status: complete` or no active task: skip