agent-skills-gap
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted user-provided logs and prompts, creating a surface for indirect prompt injection.
- Ingestion points: The
cli.jsscript reads arbitrary text or JSONL files provided by the user viafs.readFileSyncfor processing by the analyzer. - Boundary markers: The
analyzer.jsscript attempts to differentiate between user and assistant roles using regex patterns and JSON structure parsing to limit context confusion. - Capability inventory: The skill's capabilities are limited to local file system operations (reading inputs and writing JSON/SVG results) and directory creation. No network capabilities are present in the provided scripts.
- Sanitization: The
cli.jsscript includes anescfunction that performs basic HTML/SVG entity encoding on category labels and levels before they are interpolated into the SVG share card. - [COMMAND_EXECUTION]: The skill requires the agent to execute a local Node.js command-line interface (
cli.js). The instructions specify a fixed command structure and provide guidance on handling sensitive input via file paths rather than shell arguments to reduce the risk of accidental shell injection.
Audit Metadata