The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides patterns for storing and retrieving facts from user interactions, creating an attack surface where malicious instructions could be persisted in long-term memory.
Ingestion points: The IntegratedMemorySystem.store_fact method in scripts/memory_store.py and the VectorStore.add method in references/implementation.md ingest data directly into storage.
Boundary markers: The implementation uses basic markdown headers (## Relevant Memories) to delimit retrieved content when building context, which lacks robust isolation to prevent an LLM from following embedded instructions.
Capability inventory: The skill provides tools for state management but does not include capabilities for shell execution or network access.
Sanitization: There is no evidence of input validation, filtering, or escaping for the data stored in the memory systems.- [EXTERNAL_DOWNLOADS]: The skill's implementation scripts and technical references depend on the numpy library for vector operations.- [METADATA_POISONING]: There is an inconsistency between the skill's author context and the metadata provided in plugin.json.

memory-systems