The Agent Skills Directory

Data Exposure & Exfiltration (HIGH): The skill provides a 'Full home directory browser' via API endpoints like /api/files/read?path=<path>. This allows unauthorized users on the network or Tailscale mesh to read sensitive files such as SSH keys, cloud provider credentials, and environment variables from the host machine.
Persistence Mechanisms (HIGH): The skill includes a management script to install the server as a launchd service (./manage.sh service install), ensuring the application and its associated risks persist across system reboots and crashes.
Command Execution (MEDIUM): The ./manage.sh script performs high-privilege operations including service installation, dependency management, and build processes, which execute arbitrary commands on the host.
Indirect Prompt Injection (LOW): The skill ingests data from local files and Obsidian vaults which are untrusted sources. Malicious instructions inside these files could influence the integrated Claude Code chat behavior.
Ingestion points: Files accessed via /api/files/read and notes via /api/knowledge/note (SKILL.md).
Boundary markers: None identified in documentation.
Capability inventory: Arbitrary file read, local network listening, WebSocket communication (SKILL.md).
Sanitization: None documented.

MobileApp