My New Claude Code Process And Flow

Here is my current best process for Claude Code:

• Scoping. Tickets get created, then a high level grilling session.

• Kickoff. Ticket gets split up. Plan, context, test plan, smoke test plan. Another grilling session against the plan more low level.

• Implementation. Claude does what it’s good at, with more hooks than normal.

• Self-verification. Claude has to verify its own work before it’s allowed to tell me it’s done. MCPs verify against something real, then self-review runs against a critical rules file (recently I’ve been playing with these files).

• Draft PR. I review better in GitHub personally. verify-work runs the heavier rules here, the ones I don’t want used on every implementation step.

• Ready PR. pr-ready checks the PR template, runs e2e tests, attaches MCP screenshots, runs comment review and code review and architecture review. It also tells me what to smoke test by hand and what to screen record.

• Feedback and learning. Where the log gets processed.

I highly recommend adding this feedback file to your process:

Then review it every week or two.

I used Parallel research and Claude research to help me build this flow.

Every week, write down your process and flow.

Think through what you like and don’t.

Note it all down, a big context dump.

Then talk through with AI to improve or expand upon that.

Also take your note and send Claude on a Research prompt to improve upon that idea.