The Agent Skills Directory

[Data Exposure & Exfiltration] (SAFE): The skill interacts with a local Obsidian vault through MCP tools. No network operations or external data transfers were detected. Sensitive file access is restricted to the user-defined vault.
[Prompt Injection] (SAFE): The instructions are focused on structured note-taking and do not contain any bypass markers or attempts to override system safety protocols.
[Indirect Prompt Injection] (LOW): As a note-taking skill that both reads from and writes to a local database (the vault), there is an inherent surface for indirect prompt injection. A user could store a malicious instruction in a note that the agent later processes.
Ingestion points: User-provided idea descriptions are stored in Markdown files.
Boundary markers: The templates do not use specific delimiters to isolate user-generated content from agent instructions during retrieval.
Capability inventory: The skill uses obsidian_get_file_contents (read), obsidian_append_content (write), and obsidian_delete_file (delete).
Sanitization: Filenames are sanitized via slug generation, but note content is stored as-is without escaping.
[Command Execution] (SAFE): File operations are performed through specialized MCP tools rather than raw shell commands, significantly reducing the risk of arbitrary command execution.

idea