Internal working system

Stop briefing your AI from scratch

Pick up the work without repeating the whole brief.

A working setup for Claude that connects project context, reusable instructions and reviewed lessons, with checks before changes are made.

Workflow design, context architecture and AI-assisted implementation · Ongoing internal development · AI-assisted project work

How work resumes

Retrieve the project context
A task leads to the relevant brief, decisions and current status sources.
Review the proposed action
The assistant’s next step is checked against the scope and required validation.
Propose a reusable lesson
Useful learning returns through a human review step before being shared.

Simplified explanation of the internal workflow. Client information and operational details are excluded.

The problem

Work spans many projects and sessions. Decisions and useful lessons can remain buried in an old conversation or a single repository. Starting a new AI session then means explaining the same background again, with a risk that an outdated note is treated as the current plan.

The outcome

An assistant can retrieve the relevant project context and prior lessons, while task records make it possible to resume work across sessions. The system is used in my own AI-assisted work. Adapting it for a client requires scoping the team’s tools, information and permissions; it is not a packaged installation or a claim of measured client savings.

My contribution

I built a context hub that routes a question to the relevant project sources and approved lessons. Project status, reusable knowledge and the starter templates have separate owners. A reusable Claude setup adds task checkpoints, specialised instructions, review steps and validation hooks. Proposed lessons are reviewed by a person before becoming shared knowledge.

The specifics

Relevant sources are retrieved for the task, with pointers back to the original information
Project status stays separate from reusable lessons and starter templates
New shared lessons pass through human review

How it works

The context router returns source references and distinguishes stored status from information needing a live check
Reusable project templates include resumable task records and validation steps
A lesson moves from a proposed note to durable knowledge only after a human decision

Safeguards and quality

Context, knowledge and reusable project setup remain distinct concerns
Review and validation hooks support the instructions given to the assistant
Company-specific governance remains separate from the personal context system

Stack

Claude Code
Python
Reusable skills
Markdown
Git

Have a process like this one?

Email Marwane Send a project brief

aityahia@meetmarwane.com

More work