The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides a structured memory system that ingests and retrieves data, creating a potential surface for indirect prompt injection.
Ingestion points: Data is ingested into memory/ontology/graph.jsonl via CLI commands and manual entry by the agent.
Boundary markers: The skill does not implement specific boundary markers or delimiters to isolate stored entity properties from being interpreted as instructions when read by the agent.
Capability inventory: The skill possesses file-write capabilities within the workspace and facilitates agent planning through graph transformations.
Sanitization: While the skill uses yaml.safe_load and json.loads to prevent data-level attacks, it does not sanitize the semantic content of entity properties (e.g., notes, descriptions) to prevent instructions embedded in that data from influencing the agent.
[EXTERNAL_DOWNLOADS]: The management script scripts/ontology.py requires the PyYAML library for processing schema files. This is a well-known and standard library for YAML handling.

ontology