coreline-ai Parallel Dev
Overview
Turn one implementation goal into safe, independently executable workstreams. The priority is maximum concurrency without hidden file conflicts.
Workflow
- Confirm goal, non-goals, active local changes, and existing plans.
- Map likely touch points with
rg --files,rg, and docs. - Put shared contracts/types/interfaces into a small foundation workstream.
- Split leaf work by path ownership wherever possible.
- For each workstream, define objective, deliverables, owned paths, non-owned paths, dependencies, tests, and done criteria.
- Decide whether each slice is safe to run in parallel or must be sequential.
- Merge foundation first, then leaf workstreams, then shared wiring/docs.
- Run targeted tests per workstream and full verification after integration.
Sub-Agent Assignment Rules
When using sub-agents:
- Give each worker explicit file ownership and non-ownership.
- Tell workers they are not alone in the codebase.
- Tell workers not to revert or overwrite edits made by others.
- Prefer bounded implementation tasks over broad exploration.
- Keep final integration and full verification in the main agent.
Output Shape
- Parallel workstreams
- Ownership boundaries and contracts
- Integration/merge order
- Tests and smoke checks
- Risks and required coordination
Read references/parallel-templates.md when a structured workstream card or sub-agent prompt is useful.