@deepseek-ai/schemastery
@deepseek-ai/schemastery is a shared library inside a DeepSeek Harness profile: schema validation for plugin configuration.
Currently 3.19.0. It is not a plugin row: it has no panel to install and should not be removed on its own, because other packages are built on it.
What it is
A plugin declares its options once; the settings UI renders the form from that declaration, and a value outside it is rejected before the plugin starts.
Seven of the 16 profile packages depend on it directly, which is what makes the settings pages of different plugins look and behave the same.
It arrives with a profile, not as an install
7 of the 16 profile packages depend on it directly, so a profile that boots already has it.
The dependency line is in $DSH_HOME/profiles/<name>/package.json. In --dump-config, a plugin's settings appear as the options object that this schema validated.
"@deepseek-ai/schemastery": "3.19.0"
# $DSH_HOME/profiles/<name>/package.jsonWho uses it
14 community plugins declare it as a peer; each page shows what it patches.
Version and engines
npm publishes 3.19.0 (2026-11-04), up from 3.18.6-alpha.2 the day before. No engines.dsh: it is a validator, not a plugin. Its own dependencies are @deepseek-ai/cosmokit and @standard-schema/spec.
Common questions
Do I configure it directly?
No. You configure the plugin, and the plugin declares its options as a schema; this package is what rejects a bad value. Editing its own config would mean editing a validator, which changes nothing about what the plugin accepts.
What does a schema rejection look like?
The plugin does not start with the value it was given: the boot prints the requirement from the schema, and the profile continues without that row. It is the same class of failure as an engines.dsh mismatch, reported at a different layer.