Product features: skills and agents

Retrieval points: Skills Center can import / export a skill pack (standard or with history); the agent marketplace can save as a template and review to list; in chat you can type / to pick a skill; the assistant usually reads it first then runs it.

Skills Center

• Entry: Go to Skills Center (select a workspace first).
• Import / export a skill pack: see the dedicated section below (New ▾ → Import; export zip from detail).
• The list shows skills available in the current workspace; during chat the model may also cite some of them; after send the UI may say “which skills this round referenced” (see “How to use skills in chat” below).
• Top-right of the list: Refresh to reload current workspace skills (including skills other members just froze or installed).
• Conversation-derived skills (usually marked by conversation source): tap Detail for body and version history, and if needed roll back the current published version to a historical one.
• Detail page common sections: Edit (change the body), Improve (revise with a chat assistant; line-level diff before accept), trial recall, version history.
• You can create a blank skill or pick templates such as API call or Data export; or Describe to create (natural language draft, may include scripts and reference docs). Writing requirements and full templates: How to write workspace skills.
• From Messages tell the help agent or a work agent “help me generate a skill”, or after saving a connection in Data integration → Data connections tap Generate skill to write predefined-query purposes and parameters into a workspace skill. How query-type skills split with predefined queries: Data connections and predefined queries and How to write workspace skills (§8.6).

Import and export a skill pack

For migrating skills across workspaces, or backup then import. How to import a skill, how to export a skill:

On import, if this workspace already has the same id: Overwrite existing or Copy (new id).

If the skill depends on predefined queries, the export pack tries to include query definitions (not database accounts). On import, if the target workspace lacks a same-named query, a workspace admin can write them onto a data connection, or import the skill first and fill Data integration. See Data connections and predefined queries.

Install from a directory (optional)

If the platform enabled an external skill directory, Install from directory appears beside the list: browse public skill packs and install into this workspace’s Skills Center (same as importing a zip into this workspace; no out-of-pack install scripts). If not enabled, the entry is unavailable; ops notes: Server configuration.

How to use skills in chat

• In Messages type / to pick a workspace skill, or say a trigger phrase (from the skill’s “When to use”). When unnamed, the system only wakes skills clearly related to this sentence — common words like “check” or “query” will not drag in unrelated skills.
• Before fetching data, writing files, or running scripts, the assistant usually reads the skill body; the UI may show “reading skill…”.
• After this round, Messages may say “the model referenced these workspace skills this round”, marked e.g. (read), (named) (you named it with /).
• To freeze “a report produced by a skill in chat” into an app UI, see Workspace apps · Run a skill report first.

Agents: marketplace and mine

• Entry: Go to Agent marketplace (select a workspace first).
• Left: switch My agents and Agent marketplace.

Agent marketplace (three tabs)

Save a mature agent as a listing (on My listings):

1. Pick a configured assistant → Save as template (brings persona, operating rules, skill binds, and avatar; not personal memory).
2. Optionally also list on the marketplace (free reuse for others), or save as draft only.
3. Draft path: Draft → Submit for review → List. Free templates usually auto-pass after submit; you can unlist then delete.

Common statuses: draft, pending review, passed/listed, rejected, unlisted.

My agents

Manage instances you created — rename, edit notes and config, maintain the knowledge base, configure website embed, etc. Website embed is in that assistant’s Manage, same place as knowledge and human support. Before copying the embed script you can pick: corner button only, a greeting the first time, or expand chat as soon as the page opens.

Model (My agents → Manage → Persona → left “Model”):

• Default model: default Auto (pick a suitable one for this task); or name one the platform has listed. Open the comparison table: suite composite (who is stronger at getting things done, max 5), capabilities and list price. After you pick, notes follow below.
• If it says it cannot connect to the model router: only Auto for now. Start the router, then tap Retry.
• Temperature: range 0–2, default 0.2. Lower = more stable, more by-the-book; higher = more varied wording, also easier to drift. At create you can say “model temperature 1.2”; or tell the current work agent “set temperature to 0.1”, or tell the help agent “set CRM temperature to 0.1”.

- 0–0.2: getting things done, filling forms, querying, following policy - 0.3–0.6: everyday Q&A, writing email, summarizing minutes - 0.7–1.0: brainstorm, several copy variants, another wording - 1.1–2.0: deliberately novel writing; may not follow rules — use carefully

• Show and allow changing in chat: on by default. Off: no picker in the Messages composer; chat always uses the default model above.
• The help agent has no manage page; you can still change model in chat (choice stored on this device); the help agent has no separate temperature.

How to start chatting after create

1. Tell the help agent “help me create an agent named…” (with a workspace selected it writes to the current workspace). For creative wording, add “model temperature 1.2”. You can also create an instance in the marketplace, or finish config in My agents.
2. Return to Messages and select that agent on the left.
3. Send the first message; the system creates the conversation automatically.

Available data and skills

The same workspace may have several data connections and many workspace skills. If you want one agent to query only certain databases, or prefer certain skills, configure My agents → Manage → Overview → Available data and skills (tap Settings).

Notes

• Data connections and skills can be specified separately; ticking only one leaves the other on workspace default.
• Skills live in the workspace Skills Center (team-shared). This config is which skills this agent may use or prefer in chat, not moving skills into a private folder.
• Skills frozen from chat: new skills write to the workspace Skills Center; if this agent ticked Specify skills, the system auto-adds the new skill to this agent’s available list, so “frozen but this agent cannot use it” does not happen (especially with Only the resources above).
• The built-in help agent does not support this config.

For query scenes, often combine with Data resources / Who can use and Generate skill in Data connections and predefined queries.

Agent knowledge base

• From a My agents card enter Knowledge base to maintain Markdown used only by that assistant.
• After changing the body, Generate index with AI, or chat retrieval may still use the old catalog.
• Bulk migrate: top of the knowledge editor upload a folder, upload an archive (zip/rar, auto-extracted), or download an archive; import/export apply to the current browse location (whole tree at root). Only .md and index.json; after import Generate index with AI.

Agent memory

• Memory is not the knowledge base: memory stores your preferences, confirmed decisions, project milestones across conversations; the knowledge base stores role and business notes.
• In Messages tap Remember on a message; on the agent Memory page view and edit entries and the user profile.
• Full notes: Agent memory.
• How persona / operating rules / memory / knowledge split: Practice guides.
• Vague phrasing: extra round or not (ask the model when vague / only act on clear requests): default extra round when vague; change on Persona, or say “only act on clear requests” in Messages. Generating or revising a document, writing a file, is not treated as creating an agent or app. See Workspace apps.

Agent memos

• Memos are not memory: they store employee files, leave, expenses, etc.; when needed say “look up memos / look up leave”.
• On the agent Memos page view by tag and date; full notes: Agent memos.

Scheduled tasks

• You can configure on-time reminders or auto-runs for My agents (run one round at the time and notify you).
• Entry: Scheduled tasks, or the agent detail Schedule page.
• Full notes: Scheduled tasks.

Human support

Agents that answer or serve externally can enable human support: in My agents → Manage → Human support name local colleagues, or authorize a support group. In chat users tap Human support to hand off; support takes tickets at the support desk. See Human support.

Operating rules (AGENTS.md)

• Agents that handle “last N days” or call business APIs in steps should maintain AGENTS.md (operating rules) at runtime, separate from persona SOUL.
• The platform injects current time automatically; how to compute dates and how to query is configured in that agent’s operating rules.
• Writing and templates: Agent operating rules (AGENTS.md).

How to choose: skill or agent

• Need a one-off or tool-like capability, or to see skills already in the workspace → prefer Skills Center.
• Need a long-term role, dedicated knowledge, website embed → prefer an agent (My agents / marketplace).

Related docs

• Data connections and predefined queries — queries, narrowing several connections, Generate skill from a data connection, predefined queries traveling with the pack
• How to write workspace skills — writing requirements and query / HTML interactive-page templates
• Interactive HTML report scripts — ECharts / Mermaid / Leaflet and sample skills
• What belongs in a skill (allow-list) — no extras when generating/updating
• Workspace apps — freeze a skill report into an app