Skip to main content
CollabaseAI picks its own way to an answer. You do not choose a skill from a menu: you ask in your own words, and it works out which of the skills below it needs, often several in a row. This page exists so you know what is in reach. While it works, small cards appear in the chat naming what it is doing. The headings below match those cards. Everything here is scoped to your permissions. A skill that cannot find something is often telling you that the thing exists but is not yours to see. Nine of these skills change something, and each one stops for your approval before it runs. You get a card naming exactly what it would do, with Approve and Reject. Nothing is written until you approve, and rejecting changes nothing. See Before it changes anything.

Finding your way around

Listing spaces

Names the spaces you can open. CollabaseAI runs this on its own before it uses a space name you gave it, so it never acts on a space that does not exist.
Ask: “Which spaces do I have access to?” You get: the list of spaces you can open, by name.

Listing projects

Names the projects you can open, and the space each one belongs to.
Ask: “What projects are in the Marketing space?” You get: the projects in that space. If the space name does not match anything, it lists the spaces that do exist instead of guessing.

Searching by meaning

Searching content

Searches the text of pages, work-item descriptions and comments by what it means rather than by what it is called. This is the skill that answers a question when nothing is titled after it, and it is the only one that reaches comments, where the reasoning behind a decision usually sits. It needs semantic search to be switched on by an administrator. When it is off, CollabaseAI says so instead of reporting that nothing was found.
Ask: “What did we decide about the onboarding process?” or “Why did we drop the second payment provider?” You get: the passages that answer the question, each with a link to the page or work item it came from. A comment is only ever as reachable as the thing it was written on, so results stay inside what you may open.

Work items

Searching work items

Searches work items across every project you can read. It can narrow by words in the title, by project, by space, by whether the item is assigned to you, and by whether it is still to do, in progress, done or closed.
Ask: “Show me my open work items” or “Which items in Website Relaunch mention the footer?” You get: a list with the key, title, status and assignee of each item, each one a link.

Reading work item

Opens one item and reads everything on it: status, type, priority, assignee, group, labels, start and due date, story points, parent and subtasks, custom fields, and the description.
Ask: “What is on PROJ-42?” You get: the full item, in prose, with a link to it. You can name it by key or by words from its title.

Work item history

Lists what changed on one item and who changed it, newest first.
Ask: “Who moved PROJ-42 and when?” You get: the last 20 changes, each with the date and the person.

Creating work item

Creates a new item in a project you are a member of. It searches for near-duplicates first and shows you what it found, then puts the new item up for approval before it writes anything.
Ask: “Create a work item in Website Relaunch: the footer links break on mobile” You get: an approval card describing the item, and once you approve it, the new item with its key and a link. The item is assigned to you and starts at the first status of the project’s workflow.

Updating work item

Changes an existing item: title, priority, assignee, start or due date, story points, or the description. The assignee has to be a member of that item’s project, named the way you would name them.
Ask: “Set PROJ-42 to high priority and give it to Anna” You get: an approval card listing each field it would change, then the change, recorded in the item’s activity feed under your name.
Status is not part of this skill. Moving an item is its own step, below.

Checking statuses

Shows where an item can go from where it stands right now. The workflow and its conditions are already applied, so a status missing from the list is one this item cannot reach from its current position, not one that does not exist.
Ask: “Where can I move PROJ-42 from here?” You get: the statuses actually on offer. If the project runs without a governed workflow, it says so, because then the item can go anywhere.

Changing status

Moves an item to another status along the workflow transition, rather than setting the field directly. That is what makes transition rules, required fields and follow-up actions apply, exactly as they do when you drag the card on the board.
Ask: “Move PROJ-42 to In Review” You get: an approval card naming the target status, then the move. If the target status is not in the list of reachable ones, it tells you which are, instead of forcing it.

Moving work item

Moves an item to another project, in the same space or a different one. Sub-items travel with it and each gets a new key. CollabaseAI decides the status mapping itself, matching the destination column by name first and by state second, so an item never lands in a column nobody agreed to.
Ask: “Move PROJ-42 to the Platform project” You get: an approval card naming the destination and what the move will clear, then the item at its new key. The old key keeps working, so existing links do not break. You need permission to move items, and permission to create them in the destination.

Pages and whiteboards

Searching pages

Finds pages by words in their title, across every space you can read. Pages with restricted access are checked one by one, so a page you are not on the list for never appears.
Ask: “Find the pages about onboarding” You get: the matching pages with their space, status, tags and when they last changed.

Reading page

Opens one page and reads it: the full text, its status, tags, custom fields, and who last changed it. If the page has an unpublished draft, that is what it reads, because that is what you see in the editor.
Ask: “What does the Security Policy page say about passwords?” You get: an answer grounded in that page, with a link to it.

Page history

Lists the saved versions of a page with their number, date and author.
Ask: “How often has the Security Policy page been edited?” You get: the version list, newest first.

Creating page

Writes a new page, or a new whiteboard, in a space you can create in. You describe what you want; CollabaseAI builds the document.
Ask: “Create a page in the Product space with meeting notes from today’s planning: decisions, open questions, next steps” You get: an approval card, then the finished page with a link. Ask for a whiteboard instead and it draws one.

Updating page

Adds to a page you can edit. It can append to the end, put content at the top, or replace the whole document. Every write creates a version, so nothing is lost.
Ask: “Add a section on rollback to the Deployment Guide” You get: an approval card showing what it would write and where, then the new section in the page.
Before it writes a page or a whiteboard, a Checking the page format card appears. That is CollabaseAI looking up how a Collabase document is built, so the result is a real page rather than a wall of plain text. Nothing to act on.

Cycles

Listing cycles

Lists the cycles of one project with their goal, state, dates and how many items are in them.
Ask: “Which cycles are running in Website Relaunch?” You get: the project’s cycles, newest first.

Cycle review

The numbers for one cycle: committed against delivered story points, how many items were planned and how many made it, who delivered what, which work was pulled in after the cycle started, and what did not finish.
Ask: “How did the last cycle in Website Relaunch go?” You get: the review figures, with the unfinished items listed as links. Name a cycle to review that one; leave it out and you get the running cycle, or the one that finished last.

Registry

Access to the Registry is decided per schema, not per space, so CollabaseAI reaches exactly the schemas you can open and nothing beside them.

Searching the registry

Finds objects across every Registry schema you can read: assets, systems, teams, suppliers, whatever your organisation keeps records of.
Ask: “Which servers do we have in Zurich?” You get: the matching objects with their key, type and schema, each one a link. A single unscoped search looks through up to 50 object types, so name the schema or the type on a large registry.

Reading Registry object

Opens one object and reads every field value under the field’s own name, plus the objects it is linked to.
Ask: “What is on SRV-12?” You get: the object’s fields and its relations, with the linked objects named rather than listed as references. Up to 10 relations are resolved to names per object.

Registry object history

Lists what changed on one object and who changed it, newest first.
Ask: “When did the owner of SRV-12 last change?” You get: the last 20 changes with the date and the person.

Creating Registry object

Creates a new object in a schema you can write to. You name the object type and the label it should be listed under, and give field values by the field’s own name.
Ask: “Add a server called db-prod-03 to the Infrastructure registry, environment production” You get: an approval card, then the new Object with its generated key. Field names the Object Type does not carry are reported back untouched rather than silently dropped.

Updating Registry object

Renames an object, or changes any of its field values, addressed by field name.
Ask: “Set the owner of SRV-12 to the Platform team” You get: an approval card, then the change, recorded in the Object’s history.

Test management

A test project is its own container for test cases, runs and milestones, and it is separate from a work project. These skills need the View Test Results permission in the space.

Listing test projects

Names the test projects you can open, with the space each belongs to and how many test cases and runs it holds.
Ask: “Which test projects can I see?” You get: the list, so you can name the right one in the questions below.

Searching test cases

Finds test cases by words from their title, across every test project you can read.
Ask: “Which test cases cover the checkout?” You get: the matching cases with their state, priority, type, suite and the work items they cover.

Listing test runs

Lists test runs with their state, milestone, dates and result counts.
Ask: “What is currently being tested in the Web QA project?” You get: the recent runs, each with how many results passed, failed, are blocked, were skipped or are still open.

Reading a test run

Opens one run and reads its results case by case.
Ask: “What failed in the Release 2.4 regression run?” You get: every case in the run with its outcome, the tester’s comment, when it was executed and who it is assigned to, plus the totals.

Reading a test milestone

Opens one milestone: its due date, state, the runs attached to it, and the work items linked to it.
Ask: “Is the 2.4 release milestone ready to ship?” You get: the milestone with its runs and their state, and the linked work items. Work items in projects you cannot open are left out.

Discussion

Reading discussion

Reads the comments on one work item or one page.
Ask: “What was decided in the discussion on PROJ-42?” You get: a summary of the last 20 comments with who said what.

Writing a comment

Posts a comment on a work item or a page as plain text, from you. Mentions and notifications behave exactly as they do when you type it yourself.
Ask: “Comment on PROJ-42 that we are waiting for the design review” You get: an approval card with the exact wording, then the posted comment.

Planning

Planning your day

Gathers what is on your plate: work assigned to you that is overdue, due today, coming up in the next few days, and work you have already started that has no date pushing it. That last group is the pile that quietly grows. CollabaseAI collects the facts, then draws the conclusion in its answer. It reads the priorities, how late something is and whether it sits in a running cycle, and says what to do first and what can wait. The reasoning is in the answer, where you can disagree with it.
Ask: “What should I do today?” or “What is on my plate this week, and what can wait?” You get: your overdue and due work grouped, with a recommendation of where to start. The horizon is the next 7 days by default and at most 30.

Looking back

Recent activity

A digest of what happened across your work: items created, completed and changed, pages written and edited, and comments. It covers the last two weeks by default.
Ask: “What happened in the last two weeks?” or “What did I miss while I was away for a month?” You get: a grouped summary, with links. It looks back at most 90 days and returns at most 25 entries per category, so a very busy period is a sample rather than a complete record.
The digest covers work items, pages and comments. Registry objects, test runs and time entries are not in it yet.

What you have open

Fetching context

Reloads the live state of the work item, page or Registry object named in the pill under the panel header. CollabaseAI already has that context when you open the panel; this refreshes it after something has changed.
Ask: “What changed here just now?” while a work item is open. You get: an answer read from the item as it stands this second, not as it was when you opened the chat.
Last modified on July 30, 2026