ai-memory-durable-pages
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to capture, structure, and store project knowledge, session narratives, and ADRs (Architectural Decision Records) into a durable wiki that is read by agents in future sessions.
- Ingestion points: The skill uses
memory_write_pageto ingest user-supplied notes, annotations, and "raw observations" from existing sessions (as described in the 'Writing a session's pages yourself' section inSKILL.md). - Boundary markers: The instructions specify using Markdown
# H1headers for titles but do not provide delimiters or "ignore embedded instructions" warnings for the body of the memory pages. - Capability inventory: The skill has tools for writing (
memory_write_page), deleting (memory_delete_page), and restoring (memory_restore_page) persistent memory across projects and global profiles. - Sanitization: There are no requirements or instructions for the agent to sanitize, escape, or validate external content before interpolating it into the durable memory store. This could allow malicious instructions stored in one session to be executed by an agent in a future session when that memory is retrieved as context.
Audit Metadata