kaizenlisted
Install: claude install-skill MarkBovee/nebu-skills
# Nebu Kaizen
Default to steady progress: inspect, create, test, review, continue. If the next logical step is clear, keep going without pausing for routine check-ins.
## Choose the mode
- **Direct:** known files, tight coupling, fast iteration, nuanced judgment
- **Delegate:** independent research, parallelizable subtasks, noisy command runs, or specialized review
- **Batch:** related reads, searches, and edits that can be done safely together
## Cheap-first escalation
1. Start bounded mechanical chores on the smallest viable agent or subagent.
2. Validate the result before widening context.
3. Escalate to default agent only if scope grows beyond the original bounded task.
4. Escalate to high or xhigh only for cross-cutting, analysis-heavy, or repeatedly failing work.
## Core loop
1. Inspect the next boundary that matters.
2. Create the smallest coherent improvement.
3. Test it with the fastest trustworthy proof.
4. Review it for clarity, safety, consistency, and scope.
5. Continue unless a real blocker or decision point appears.
## Default rules
1. Do not stop after every milestone when the next step is already clear.
2. Ask only when the answer changes scope, product behavior, architecture, safety, or acceptance.
3. Prefer preventing mistakes early with types, validation, guards, and simpler control flow.
4. Follow existing repo patterns before inventing new ones. Build only what the current requirement needs.
5. After code changes, do a proportional review pass.