The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the AI_MEMORY/L2_行为层/ directory to perform behavioral analysis. While it lacks explicit boundary markers or sanitization for this data, the capability inventory is limited to local file read and write operations within the designated memory directories. There is no evidence of network tools, shell access, or other dangerous capabilities that could be exploited via injection.
[SAFE]: No obfuscation, remote code execution, or persistence mechanisms were detected. The skill's design respects the integrity of the core identity layer by requiring manual user updates for L4 files, and all external interactions are confined to the intended memory management scope.

mem-monthly