Evernote Official Assistant

The problem

In engineering workflows, Evernote is often used to capture documents, clips, tags, and notebooks. When an AI agent operates it, three failure modes are common: local files or mock data replace real API results; guessed tokens make authorization unreliable; and destructive operations such as updates, moves, or tag changes are batched without confirmation. Search, list, and detail output can also leak raw fields or body text.

How it works

This skill treats Evernote as a remote API surface. It detects bash/zsh versus PowerShell, loads the saved token, and stops for authorization if the token is missing. Each supported action maps to an API operation: create note, update note, move note, change tags, web clip, create notebook, create tag, search notes, list notes/notebooks/tags, and fetch note detail. Batch tasks are split into atomic operations and executed in dependency order. When targets must be inferred from search or list output, the skill asks for confirmation before modifying them.

Field rules are strict. Tag updates send the final full tagNames list; clearing tags requires clearTags:true; replacing note content requires explicit confirmation. Search uses creation-time filters with startTime and endTime as UTC milliseconds. User-facing replies summarize results in natural language instead of exposing raw JSON, code, status, or message fields.

Boundaries

It is suitable for real Evernote knowledge management and automated note workflows. It is not suitable for unsupported actions such as querying by update time, bulk deletion, or export. For any unconfirmed search, list, move, or tag change, confirmation should happen before execution.