local-server

Warn

Audited by Gen Agent Trust Hub on Mar 27, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes python3 execution/local_server.py to start a local FastAPI server and uses cloudflared tunnel to establish network connectivity.
  • [DATA_EXFILTRATION]: The instructions facilitate exposing a local environment to the public internet via a Cloudflare tunnel. The documentation explicitly notes that the server has full access to local files and credentials (via .env files), creating a risk of data exposure if the endpoints are accessed by unauthorized parties.
  • [EXTERNAL_DOWNLOADS]: The skill utilizes the cloudflared utility, a well-known service provided by Cloudflare, to establish a tunnel for external webhook access.
  • [PROMPT_INJECTION]: The skill creates an attack surface for indirect prompt injection by exposing endpoints like /directive and /general-agent to external network traffic.
  • Ingestion points: External HTTP requests to the FastAPI endpoints.
  • Boundary markers: None identified in the provided instructions.
  • Capability inventory: The skill is configured with powerful tools including Bash, Read, Write, and Edit, which are accessible to the orchestrator running on the local server.
  • Sanitization: No input validation or sanitization logic is described for the processed directives.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 27, 2026, 12:41 AM
Security Audit — agent-trust-hub — local-server