The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill downloads a self-contained viewer tarball from a remote GitHub URL and executes it directly using npx --yes. This allows the execution of arbitrary code fetched from the internet at runtime.
[COMMAND_EXECUTION]: The skill directly interpolates user-provided $ARGUMENTS into a Bash script variable (PROJECT_ARG="$ARGUMENTS") without sanitization. This is a command injection vector that could allow a malicious project path to execute unintended shell commands.
[COMMAND_EXECUTION]: The skill performs automated package management and build operations using pnpm install and pnpm build within the plugin's directory.
[INDIRECT_PROMPT_INJECTION]: The skill reads from a local knowledge-graph.json file which could be manipulated by a malicious repository to influence the dashboard's output or potentially exploit the visualization tool.
Ingestion points: knowledge-graph.json, package.json
Boundary markers: None used when reading JSON data.
Capability inventory: Subprocess execution via npx, pnpm, and node; file-system access via Bash.
Sanitization: No validation or sanitization is performed on the content of the knowledge graph before it is passed to the dashboard server.

understand-dashboard