The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from project files such as Java source code, React components (TSX), and build configuration files. This establishes an attack surface where code being reviewed could contain instructions designed to bias the analysis report. However, the skill design as a 'report-only' tool without code execution or file-write capabilities significantly mitigates this risk. 1. Ingestion points: Java, TSX, CSS, pom.xml, and build.gradle files. 2. Boundary markers: No explicit markers are used when processing file content. 3. Capability inventory: Uses MCP tools for documentation lookup; no file-system write or shell execution capabilities. 4. Sanitization: Relies on review logic rather than explicit sanitization.
[DYNAMIC_EXECUTION]: The skill performs dynamic task orchestration by generating instructional 'contracts' for sub-agents or processing batches of files. This involves assembling prompt context at runtime based on project metadata like version and theme. This is a standard and safe methodology for multi-file analysis tasks.
[SAFE]: All external resource calls are directed to the Vaadin MCP server, which is consistent with the skill's purpose and its author context. No credential harvesting, data exfiltration, or malicious persistence mechanisms were detected.

vaadin-usage-review