The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process large volumes of untrusted data from external sources including Obsidian vaults, Notion exports, and JSON/CSV files. This data could contain malicious instructions meant to influence the agent's behavior during or after the migration process.
Ingestion points: Processes external directories, Notion export files, and structured JSON/Roam exports (SKILL.md).
Boundary markers: The skill explicitly references skills/conventions/untrusted-content.md and warns the agent that imported instructions must be treated as data, never as new operating authority.
Capability inventory: Accesses tools such as put_page, search, and sync_brain, and executes shell-based CLI commands for data inspection and synchronization.
Sanitization: Relies on verification receipts, health checks, and spot-checking, although these are focused on data integrity rather than prompt sanitization.
[COMMAND_EXECUTION]: The skill provides instructions to use a specific CLI tool (gbrain) to perform data operations.
Evidence: Commands like gbrain sources inspect <path>, gbrain sources connect <path>, and gbrain extract links are used to manage the migration workflow.
Mitigation: The instructions explicitly require non-interactive confirmation (the --yes flag) and operator approval of previews before critical connection steps are taken.

migrate