Tutorials for DeepSeek Harness plugins

Run npx @deepseek-ai/dsh web, confirm the local URL, and check the version reported by dsh --version before adding plugins.

Use dsh plugin --profile web add <package>, then verify it appears under Settings → Plugin Market and answer any permission prompts.

Why the Web UI, model adapters and tools all share one plugin model, and what that implies for which extension points are available to you.

A tour of index.json format 1: categories, labels, sort keys, plugin versions, npm sources and engines.dsh ranges.

What each profile loads, how to keep experimental plugins out of your daily profile, and when to create a dedicated profile.

Where plugin settings live, how to override defaults per project, and how to check what a plugin actually received at startup.

Pin versions, read the changelog, snapshot your config, then upgrade one plugin at a time so failures stay isolated.

Check every installed plugin's engines.dsh range first, update the core, then reapply plugins that declare a compatible range.

Remove it from the profile, clear leftover config keys, and check that no other plugin still depends on it.

The four usual causes: wrong profile, missing restart, permission not granted, or a version outside the engines.dsh range.

Find the plugin log, separate a resolution failure from a runtime throw, and identify the offending plugin when several fail at once.

Scaffold a package, export apply(ctx), register one tool, then test it locally before wiring up npm publishing.