run-remote-dev-server
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill follows security best practices for remote access and development. It explicitly instructs the agent to avoid common security pitfalls such as binding services to all interfaces (0.0.0.0), disabling firewalls, or using public tunnels (Tailscale Funnel). It emphasizes maintaining private previews through Tailscale Serve and ensures that credentials like private keys or session tokens are never stored or transmitted insecurely.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external repository data (README, project files, repository instructions) which represents a potential injection surface. However, the skill incorporates strong mitigations by instructing the agent to independently validate all commands, maintain strict action boundaries, and ignore any directions in external files that conflict with user requests or security policies. 1. Ingestion points: Reads repository instructions, project files, and READMEs from remote hosts (specified in SKILL.md). 2. Boundary markers: Explicitly commands the agent to ignore embedded directions that conflict with safety boundaries or the user's request. 3. Capability inventory: Includes remote command execution via SSH and network route management via Tailscale. 4. Sanitization: Instructs the agent to independently validate task-relevant commands and verify facts directly rather than relying on potentially poisoned documentation.
Audit Metadata