drawio-skill

Hardening Guide: drawio-skill

Least-privilege permissions

• Restrict file access: Only grant the skill permission to read infrastructure files it needs; block access to sensitive folders, credential vaults, and unrelated systems.
• Control output destinations: Limit where diagrams can be saved or exported (e.g., only to internal storage, not cloud services or public URLs).
• Separate user roles: Ensure only authorized team members can run the skill and view generated diagrams; prevent junior staff from accessing sensitive infrastructure topology.
• Rotate credentials before parsing: If possible, temporarily revoke or mask secrets in Terraform/Kubernetes files before feeding them to the skill.

Network isolation

• Keep processing local: Ensure diagram generation and rendering stays within your private network; configure the skill to never send data to external draw.io cloud services without explicit approval.
• Block outbound connections: Use firewalls to prevent the skill from reaching unfamiliar APIs, file-sharing services, or cloud storage without prior authorization.
• Validate export services: If diagrams must be exported (to PowerPoint, HTML, etc.), verify the export process stays on-premises or uses only trusted, vetted vendors.
• Segment from production: Run the skill on isolated systems separate from live infrastructure; prevent it from directly querying production databases.

Monitoring

• Log all diagram generation: Record what infrastructure files were parsed, who requested the diagram, and where outputs were sent.
• Audit input/output content: Scan generated diagrams for unexpected service names, database credentials, or sensitive IP addresses; flag unusual patterns.
• Track file access: Monitor which users access infrastructure source files before and after skill execution to detect misuse.
• Alert on suspicious parsing: Set up warnings if the skill encounters malformed or unusually large files, which may indicate attack attempts.

Red flags to watch for

• Diagrams containing plaintext secrets: If exported diagrams include API keys, passwords, or connection strings, immediately revoke those credentials.
• Unexpected external connections: Any attempt by the skill to contact cloud services, CDNs, or unknown domains outside your approved vendor list.
• Permission escalation requests: If the skill suddenly demands access to credential stores, admin accounts, or production systems it didn't need before.
• Garbled or suspicious diagram logic: If the skill generates components or connections that don't match your actual infrastructure, this may indicate prompt injection or parser abuse.