migrating-memory

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION] (HIGH): Indirect Prompt Injection vulnerability via memory inheritance. The skill copies files from a source agent into the destination agent's /system/ memory directory. Because 'system' blocks are typically loaded directly into the agent's active context/prompt, this allows for the injection of instructions from an external source.
  • Ingestion points: Files are read from /tmp/letta-memfs-<source-agent-id> and written to ~/.letta/agents/$LETTA_AGENT_ID/memory/system/ (SKILL.md).
  • Boundary markers: None present; the skill treats all migrated files as trusted data.
  • Capability inventory: Use of cp for file system modification and letta memfs sync for updating the agent's operational state based on those files.
  • Sanitization: None; the skill does not validate or sanitize the content of the memory blocks being migrated.
  • [COMMAND_EXECUTION] (LOW): The skill utilizes standard shell commands (cp, letta) to perform local file operations and interact with the Letta CLI. While these are executed in the local environment, they are used as intended for the skill's purpose.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 16, 2026, 09:12 PM