The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from a user-provided or generated JSON file, creating a surface for indirect prompt injection if that data contains malicious instructions.
Ingestion points: Metadata and graph structure are read from /docs/learning-graph/learning-graph.json as specified in the Installation Workflow and assets/script.js.
Boundary markers: There are no explicit markers or instructions to the agent to sanitize or ignore embedded instructions within the JSON file content.
Capability inventory: The skill performs file system modifications (mkdir, cp), executes shell-based Python snippets, and runs browser-side JavaScript to render visualizations.
Sanitization: The viewer's JavaScript implementation utilizes textContent for rendering node labels and statistics, which prevents basic cross-site scripting (XSS), although the initial title replacement in main.html is performed via simple string interpolation.
[EXTERNAL_DOWNLOADS]: The skill incorporates an external visualization library from a well-known service.
Evidence: The assets/main.html file includes a script tag fetching vis-network.min.js from the unpkg.com CDN.
[COMMAND_EXECUTION]: The installation process involves standard shell commands for directory management, file copying, and data extraction.
Evidence: The skill uses mkdir, cp, and python3 -c for project setup and to parse the learning graph title during the installation workflow.

install-learning-graph-viewer