Manus Commands & Prompt Reference

27 commands & parameters · 4 categories · Source: Manus API Reference (open.manus.ai)

Manus is an agent rather than a chat model, and its API shape reflects that: you create a task, the agent plans and executes it asynchronously, and you poll or stream messages until it delivers a result. The typed surface is correspondingly different from a chat completions endpoint - task-level settings like agent_profile, locale, and interactive_mode sit alongside a message object whose fields decide which skills, connectors, and prior tasks the agent may reach for.

Three message fields do most of the steering. enable_skills loads a skill set for the task, force_skills guarantees the agent actually invokes them, and task_references (up to 20 bare 22-character task IDs) lets the agent browse earlier work. Attachments arrive as ContentPart items, and which field you use matters: file_id for something pre-uploaded, file_url for a URL Manus downloads (20 MB cap), or file_data for inline base64 (also 20 MB after decoding).

All Commands

Task Parameters (8)

Command / SyntaxWhat it doesNotes
project_idProject ID to associate this task withThe project's instruction will be automatically applied. Use project.list to get available project IDs.
localeLocale for the task output language (e.g., "en", "zh-CN", "ja")Defaults to the user's account locale setting.
interactive_modeWhen enabled, the agent may pause and ask follow-up questions if the input is insufficientWhen disabled (default), the agent proceeds with best-effort execution without asking.
hide_in_task_listWhen true, the task will not appear in the Manus webapp task listThe task is still accessible via the task_url in the response. Useful for automated/background tasks.
share_visibilityControls who can view the task"private" (default) — only the task creator can view. "team" — all team members can view. "public" — anyone with the share_url can view without authentication.
agent_profileAgent profile to use for the task"manus-1.6" (default) — standard capability. "manus-1.6-lite" — lightweight, faster responses. "manus-1.6-max" — maximum capability. Free personal accounts are downgraded to...
titleCustom title for the taskIf not provided, a title will be auto-generated based on the input message.
structured_output_schemaJSON Schema for structured output extractionWhen provided, the agent runs normally, then a post-processing step extracts a result conforming to your schema. The schema must follow the Structured Output subset of JSON Schema: all...

Message Object Fields (5)

Command / SyntaxWhat it doesNotes
message.contentString (plain text) or array of ContentPart objectsUser text is limited to approximately 5,000 estimated tokens per request; the exact character count varies by language and content. For an array, the limit applies to the combined text...
message.connectorsList of connector IDs to enable for this taskResolution order when omitted: 1) if the task belongs to a project, the project's default connectors are used; 2) otherwise, the user's default enabled connectors are used. Use...
message.enable_skillsSkill IDs to enable for this taskIf empty or omitted, loads the skills the user has enabled in their account settings. Use skill.list to retrieve available skill IDs.
message.force_skillsSkill IDs the agent must invoke during this taskForced skills are automatically available even if not listed in enable_skills.
message.task_referencesTask IDs of existing tasks to reference from this messageThe agent can then browse the conversation and files of those tasks on demand. Each entry must be a bare 22-character alphanumeric task ID — a full task URL, or a shortcut value such as...

Content Part Types & Fields (9)

Command / SyntaxWhat it doesNotes
"type": "text"Must be "text"Field of a ContentPart item inside message.content.
textThe text content of the messageAcross a ContentPart array, all text parts must contain no more than approximately 5,000 estimated tokens in total.
"type": "file"Must be "file"Field of a ContentPart item inside message.content.
file_idID of a previously uploaded fileUpload via file.upload first to get the file ID.
file_urlPublicly accessible URL to the fileThe agent will download it directly.
file_dataBase64 encoded file contentFormat: data:<mime_type>;base64,<encoded_content> (e.g., data:application/pdf;base64,JVBERi0...).
filenameDisplay name of the file including extension (e.g., "report.pdf")Required when using file_data; recommended for file_url.
mime_typeMIME type of the file (e.g., "application/pdf", "image/png")Optional; auto-detected from filename if omitted.
"type": "voice"Must be "voice"Field of a ContentPart item inside message.content.

Endpoints, Profiles & Follow-up Fields (5)

Command / SyntaxWhat it doesNotes
POST /v2/task.createCreate a new task; runs asynchronouslyPoll task.listMessages for progress. OAuth scope: create_task or manage_all_tasks.
POST /v2/task.sendMessageSend a follow-up message to an existing taskUse this instead of task.create when talking to an agent that already exists.
manus-1.6Default agent profile: standard capabilitySet as the agent_profile value on task.create or per turn on task.sendMessage.
task_idThe unique identifier of the task to send the message toSupports the shortcut agent-default-main_task for the IM agent's main task.
clear_connectorsRemoves all connectors from this task for this and subsequent turnsCombine with message.connectors to form three states: a non-empty message.connectors overrides the task's connectors with the new list; clear_connectors: true clears all connectors;...

Frequently Asked Questions

How do I continue an existing Manus task instead of starting a new one?

Use task.sendMessage with the task_id rather than task.create. It accepts the same message content formats.

What is the difference between enable_skills and force_skills?

enable_skills makes a skill available for the task; force_skills guarantees the agent invokes it. Forced skills are automatically available without also being enabled.

How do I attach a large file?

Upload it first via file.upload and pass file_id. file_url and file_data are capped at 20 MB, so large files must go through the upload path.

Every entry on this page is a literal, typeable token taken from the official documentation linked above. Command sets in AI tools change quickly - the in-product command menu is always authoritative for your installed version.

Related prompt references

Claude Code (111) · Cursor (37) · DeepSeek (22) · All AI prompt references