The Agent Skills Directory

[PROMPT_INJECTION]: The skill creates an indirect prompt injection surface by designating the target vault's local AGENTS.md file as "authoritative," stating it "overrides this skill." This allows potentially untrusted data inside the vault to redefine the agent's instructions.
Ingestion points: AGENTS.md and VAULT.md located within the root of the target markdown vault.
Boundary markers: Absent. There are no instructions to the agent to treat these files as data rather than instructions.
Capability inventory: The agent has the ability to read system config files, create directories via shell, and use various MCP tools to read/write/query files.
Sanitization: Absent.
[DATA_EXPOSURE]: To facilitate vault discovery, the skill instructs the agent to read obsidian.json from platform-specific application support directories outside the immediate workspace (e.g., ~/Library/Application Support/obsidian/ on macOS or %APPDATA%\\obsidian\\ on Windows).
[COMMAND_EXECUTION]: The skill uses local shell commands for environment checks and filesystem management.
Runs turbovault --version to verify the installation of the required CLI.
Uses mkdir -p <vault-path>/area/<name> to create new directory structures within the vault.
[EXTERNAL_DOWNLOADS]: The skill requires an external MCP server, TurboVault, and points to a third-party GitHub repository (github.com/Epistates/turbovault) for its installation and configuration.

vault-ops