start-server-routes
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documents an ingestion point for untrusted external data that lacks sanitization when paired with sensitive capabilities.\n
- Ingestion points: The splat parameter (params._splat) in the /api/file/$ route example within references/templates/rest-endpoint.md accepts arbitrary input from the URL path.\n
- Boundary markers: The instructions do not provide delimiters or warnings to ignore embedded instructions within the processed file content.\n
- Capability inventory: The example demonstrates the use of readFile(), granting a file system read capability.\n
- Sanitization: The provided code lacks path normalization or validation logic to prevent directory traversal attacks.\n- [COMMAND_EXECUTION]: The skill instructs the agent to execute a tool from the author's own infrastructure as part of its mandatory workflow.\n
- Evidence: The Agent Workflow section in SKILL.md requires running fuse-ai-pilot:sniper after modifications are made.
Audit Metadata