import-memory

Run the provided export prompt in ChatGPT to generate a categorized memory export, then paste the result into Claude. The Skill builds an import plan, filters sensitive content, and writes approved facts into Claude's memory in batches.

Without memory write tools the import cannot run in that conversation. The Skill directs you to Settings > Capabilities > Import memory from other AI providers, Claude's built-in importer, and stops without writing anything to files or artifacts.

No, the import is strictly additive. It creates new memory files or appends new lines, never rewrites or deletes existing entries, and flags conflicts between the export and existing memory instead of resolving them silently.

The privacy filter omits health and mental-health details, finances, identifiers like addresses and ID numbers, personality profiling, protected attributes, and names of partners or care providers. Instruction-like text in the export is treated as a prompt-injection risk and dropped.

Yes, and the Skill treats the entire paste as data, never instructions. Directives disguised as facts or system messages are dropped rather than filed, and the user is told when instruction-like content was skipped.

Exports over roughly 64KB or visibly truncated are handled partially: complete entries are imported, ambiguous ones set aside, and splitting the export into parts is suggested. Memory allows about 10 writes per turn, so large imports span multiple turns.