[FEATURE] /ref: per-project registry of resource pointers (where data lives and how to read it), maintained by the conversation

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

In a project I keep coming back to the same resources: a NAS share, a database, the employee list, a SharePoint folder, an API endpoint, a script. Claude has to guess or re-discover where they are and how to access them every session, and that's where errors creep in: wrong path, outdated copy, a guessed table name.

CLAUDE.md and auto memory don't solve this well. They mix instructions, notes and facts in free text. Nothing tells Claude "this is a resource, here is how to read it". They don't flag when a resource moves. And copying data into them (e.g. a list of employee names) creates stale duplicates of data that changes.

Proposed Solution

In a project I keep coming back to the same resources: a NAS share, a database, the employee list, a SharePoint folder, an API endpoint, a script. Claude has to guess or re-discover where they are and how to access them every session, and that's where errors creep in: wrong path, outdated copy, a guessed table name.

CLAUDE.md and auto memory don't solve this well. They mix instructions, notes and facts in free text. Nothing tells Claude "this is a resource, here is how to read it". They don't flag when a resource moves. And copying data into them (e.g. a list of employee names) creates stale duplicates of data that changes.

Alternative Solutions

• Today I maintain this by hand in a RESSOURCES.md file that I ask Claude to re-read every session. It works, but there's no autocomplete, no structure, and no automatic proposal when something changes.
• CLAUDE.md / auto memory: free text, no "pointer vs data" distinction, no autocomplete.
• #44598 (/pin, closed as duplicate) is about keeping content in context. This request is the opposite: keep only the address, read the content on demand.

Priority

Medium - Would be very helpful

Feature Category

Configuration and settings

Use Case Example

1. Project: an internal tool that generates weekly job-site reports.
2. Registry entries: nas-reports (SMB share path), projects-db (database + read-only user), employees (table + query), sharepoint-sites (Graph endpoint).
3. I type "update the report to include /ref employees on each site". Claude reads the current employee table instead of guessing or using a stale copy.
4. Later, the NAS share is renamed. Claude's read fails, it says "nas-reports seems to have moved, update the entry to \\new\path?", and I confirm.

Picture: the project's address book. You don't copy what's inside each house, only the address and the key.

Additional Context

Environment: Claude desktop app (Code tab), Windows, daily use for real business automation (Python/PHP tools, databases, SharePoint, NAS).