The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted repository content to build a persistent knowledge graph, which presents a surface for indirect prompt injection if the graph content is later used to generate agent responses.
Ingestion points: Repository source code files (Python, JS, Go, etc.) in the project root.
Boundary markers: None specified in the instructions for the parsing process.
Capability inventory: The skill utilizes list_graph_stats_tool and build_or_update_graph_tool for data processing.
Sanitization: No sanitization or filtering of input code comments or metadata is mentioned.

build-graph