Here are the 6 built-in slash commands of DeepSeek Harness (dsh), DeepSeek’s official agent, each with a description, when to use it and an example, based on the command docs and source code in the official repository. Type them in the input box of the Harness web UI (npx @deepseek-ai/dsh web, http://127.0.0.1:3080 by default) or desktop app; the command-line modes have no command menu. Skills you put in .dsh/skills also appear in the menu as /skill-name. Harness is still a developer preview, so commands may change between versions. If you run Claude Code on DeepSeek’s API instead, see Claude Code commands.
Modes & permissions
/plan /goal /permission
Enters or leaves plan mode: it proposes a plan first and only edits after you confirm.
Type /plan to enter plan mode; add a message to enter plan mode and send it; /plan off leaves.
/plan split the login module into its own service
Sets or shows the goal of a long-running task; Harness keeps working toward it.
Without arguments it shows the current goal; clear removes it, edit changes it, pause / resume stop and continue.
/goal make all unit tests pass and add the missing ones
Switches the permission preset (sandbox mode plus approval policy).
Presets are workspace-write (edits only inside the workspace) and danger-full-access (no restrictions, use with care); auto can be added in the config.
/permission workspace-write
Context
/compact
Compacts older conversation history to free up context space.
Use when the conversation is long or responses slow down; older turns are summarized.
/compact
Claude and Claude Code are trademarks of Anthropic, PBC. Codex and ChatGPT are trademarks of OpenAI. DeepSeek is a trademark of its respective owner. VibeSlash is an independent, unofficial open-source project and is not affiliated with, endorsed by, sponsored by or authorized by Anthropic, OpenAI or DeepSeek. These names are used only to describe compatibility; all trademarks belong to their respective owners.