The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands to interact with a code graph. It utilizes a local Python script located at .claude/scripts/code_graph to perform tasks like building the graph, tracing code flow, and analyzing the impact of changes (blast radius).
[DYNAMIC_EXECUTION]: Inline JavaScript is executed using node -e to check project configurations and manage environment setup. These scripts interface with local utility files such as .claude/hooks/lib/graph-utils.cjs to handle dependency verification and project-specific settings.
[EXTERNAL_DOWNLOADS]: The skill mentions the installation of required Python packages including tree-sitter, tree-sitter-language-pack, and networkx. These are standard libraries used for parsing source code and performing graph analysis, typically installed via pip or automated project hooks.
[INDIRECT_PROMPT_INJECTION]: The skill parses project source code to construct a knowledge graph, which introduces an indirect prompt injection surface. The instructions attempt to mitigate this by requiring structured output and evidence-based claims, but the processing of arbitrary code files is the primary ingestion point.

graph-code