LLM coding agents can drive ntropy well, since the CLI is fully scriptable, but they have to know the house rules. The repository ships an agent skill that teaches them.
Install
The quickest install is the skills CLI, which
places the skill into Claude Code, Cursor, Codex, and most other
SKILL.md-aware agents. It installs into the project by default and for
your user with -g:
npx skills add jakobwesthoff/ntropy
Or skip the tooling and copy the directory by hand. For Claude Code it
belongs at ~/.claude/skills/ntropy (every project) or
<project>/.claude/skills/ntropy (just that project):
git clone https://github.com/jakobwesthoff/ntropy.git
cp -R ntropy/skills/ntropy ~/.claude/skills/ntropy
The skill's description marks it as relevant to any task involving
ntropy, a note vault, or .ntropy-vault files, so an agent with the
skill installed picks it up on its own the moment a task touches one.
What the skill teaches
skills/ntropy/
holds a SKILL.md with the vault model, rules for agents, a do/don't
table, and a command reference. Beside it are reference docs on writing
notes, querying, vaults, views, the website export with its navigation
settings, site themes, and document themes.
The rules are the part an agent gets wrong on its own:
- Always pass
-n, and--printonnewandtoday. Without them ntropy opens a picker or the editor and blocks whenever a terminal exists; capturing the output does not prevent that. - Never hand-create files in
all-notes/. ntropy allocates the ULID, the location, and the file name; only the content is the agent's. - Author a note with
ntropy new --empty -p <title>followed byntropy write <path>, feeding the whole note on stdin. - Run
ntropy reconcileafter editing a note file directly instead of throughwrite. - Check the active vault with
ntropy infobefore changing anything, never write into the view directories, and never storeidor dates in frontmatter.