Prerequisites

Before spawning, you should have a clear picture of the work. Run /issues first to discover and understand what needs to be done.

Workflow

•Decompose the work into independent, parallelizable tasks

•Spawn a worker for each task:

bash

jig spawn <task-name> --context "<detailed context>" --auto

•
Monitor progress:
bash
```
jig ps
```
•
Review completed work:
bash
```
jig review <task-name>
```
•
Merge approved work into the current branch:
bash
```
jig merge <task-name>
```

Writing Good Context

Each --context value is the worker's entire prompt. Include:

•One-line summary of what to accomplish
•Files to modify (if known)
•Specific requirements and constraints
•Acceptance criteria — how do we know it's done?
•What NOT to do — boundaries to prevent scope creep

Example:

bash

jig spawn add-auth-middleware --context "Add JWT auth middleware to the Express API.
Modify src/middleware/auth.ts. Use jsonwebtoken package (already installed).
Protect all routes under /api/v1/ except /api/v1/health.
Return 401 with {error: 'unauthorized'} on invalid/missing token.
Do NOT modify the health endpoint or add any new packages." --auto

Rules

•Keep tasks independent — workers cannot see each other's changes
•Include enough context for the worker to be fully autonomous
•Spawn 2-4 workers at a time to avoid resource contention
•Always review before merging (jig review <name>)
•Use jig kill <name> if a worker is stuck or going off track
•After merging, check for conflicts before spawning the next batch