The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill outlines a framework for persistent memory that creates an attack surface for indirect prompt injection. 1. Ingestion points: Conversational flow and user preferences extracted from messages as described in SKILL.md. 2. Boundary markers: The framework does not describe delimiters or isolation techniques for separating stored memories from system instructions. 3. Capability inventory: Includes memory management tools such as memory_write and memory_read described in the implementation patterns section. 4. Sanitization: The documentation mentions consistency checks and quality gates but lacks specific input sanitization or instruction filtering to prevent malicious payloads from being stored or retrieved.
[NO_CODE]: The skill consists entirely of architectural guidelines and does not include any executable scripts, binaries, or shell commands.
[SAFE]: The skill includes references to legitimate AI memory frameworks (Mem0, Letta, Zep, Graphiti) and internal project documentation which serve as helpful resources for developers.

agent-memory