Prompt-Mediated Memory Portability And Extraction Protocol

Issue 60 Edition 2026-03-01 4 min read

Not accepted General

Sources: 1 • Confidence: Medium • Updated: 2026-03-08 21:21

Key takeaways

The prompt instructs the model to output every stored memory and learned context in a single code block using an entry format of "[date saved, if available] - memory content" and to avoid summarizing, grouping, or omitting entries.
Anthropic's "import your memories to Claude" feature at claude.com/import-memory is implemented as a prompt.
The prompt defines the requested scope of memories/context to include response-style instructions, personal details, projects/goals, tools/frameworks, preferences/corrections, and any other stored context, requesting verbatim preservation where possible.
The text requests a complete export of stored memories because the user is moving to another service.

The prompt instructs the model to output every stored memory and learned context in a single code block using an entry format of "[date saved, if available] - memory content" and to avoid summarizing, grouping, or omitting entries.
The prompt defines the requested scope of memories/context to include response-style instructions, personal details, projects/goals, tools/frameworks, preferences/corrections, and any other stored context, requesting verbatim preservation where possible.
The text requests a complete export of stored memories because the user is moving to another service.

Anthropic's "import your memories to Claude" feature at claude.com/import-memory is implemented as a prompt.

Does the target system actually return stored-memory records verbatim and completely when given the specified extraction prompt?
What policy and technical controls (if any) prevent full-memory dumps, sensitive personal data disclosure, or cross-user data exposure in the memory feature?
What is the precise meaning of "implemented as a prompt" for the claude.com/import-memory feature (prompt-only behavior versus prompt plus backend enforcement)?
What portion of the requested 'learned context' is actually persisted memory versus inferred profile information synthesized at response time?
Is there any direct operator/product/investor decision readthrough supported by the corpus beyond general watchfulness about memory export/import behavior?

If memory import and export are prompt-mediated, portability may be easier to ship but harder to secure, creating product and compliance implications for vendors offering memory features.
A structured extraction prompt suggests users can attempt comprehensive memory dumps, implying elevated demand for safeguards, auditability, and controlled memory portability tooling.

Vendor documentation or product behavior shows memory import or export flows are primarily prompt-driven rather than enforced by backend access controls.
Public statements, policies, or technical notes describe limits that prevent full-memory dumps and specify what categories can be exported or blocked.
Independent testing demonstrates whether the extraction prompt returns complete, verbatim stored-memory records versus partial or synthesized context.

Evidence that memory import and export are not prompt-only and are governed by strict backend controls that prevent comprehensive memory extraction.
Demonstrated inability of the extraction prompt to retrieve stored memories beyond a small, user-visible subset, indicating low real-world portability risk.
Clear vendor guarantees and enforcement that prevent sensitive data disclosure or cross-user exposure during memory export attempts.