The Agent Skills Directory

[SAFE]: The skill includes a dedicated resolve_safe_path utility in scripts/ontology.py that strictly enforces that all file operations for the graph and schema remain within the workspace root, effectively preventing directory traversal attacks.
[SAFE]: The implementation uses yaml.safe_load() for schema processing in the load_schema function, which protects against arbitrary code execution vulnerabilities during YAML deserialization.
[SAFE]: The skill documentation and schema constraints explicitly forbid the storage of secrets (passwords, tokens, API keys) directly in the graph, as defined in references/schema.md and enforced in scripts/ontology.py via forbidden_properties checks.
[SAFE]: The skill manages a knowledge graph by ingesting entity data via JSON and YAML inputs. While this creates an indirect prompt injection surface where external data enters the agent context, the skill implements robust defenses: 1. Ingestion points: External data enters through CLI arguments in scripts/ontology.py and is stored in memory/ontology/graph.jsonl. 2. Boundary markers: Data is strictly encapsulated within JSON/YAML structures. 3. Capability inventory: The skill's operations are limited to local file read/writes within the validated workspace path; it lacks network access or arbitrary command execution capabilities. 4. Sanitization: All input is parsed using standard libraries and validated against a schema that enforces type constraints and enum values.

ontology