Work packages and the AI
The AI works with work packages through MCP tools. They run the same steps as the Work Packages tab and show the same step list for approval.
When the AI uses them
The AI reads the Workflow Policy before it starts, syncs or finishes work on tickets.
- Policy active: the AI follows it and the Instructions for the AI. Work on a ticket happens on the package branch, not on the base branch.
- Policy off or missing: the AI works as usual and uses the work package tools only when you ask for them.
Example request with an active policy: "Start a package for WR-40 with WR-41 and WR-42." The AI checks the status, groups the tickets under the parent and starts the branch.
Tools
| Tool | Effect | Preview |
|---|---|---|
workflow_policy_get | Reads the policy and whether it is active. | – |
workflow_policy_set | Creates or changes the policy. Only the named fields change. | yes |
workpackage_status | Fetches and reports every open package: ahead and behind, changed files, conflicts with the base, overlaps, pull request, tickets, hints. | – |
workpackage_start | Starts a package: kind, title, parent, tickets, base branch, depends_on for a stacked package. | yes |
workpackage_switch | Checks out another package branch and parks or restores uncommitted work. | – |
workpackage_sync | Takes in the new commits of the base. Refused on conflicts. | yes |
workpackage_open_pr | Opens the pull request. Title, description and reviewers are editable in the preview. | yes |
workpackage_finish | Finishes a merged package, or abandons one. | yes |
tickets_children | Lists the tickets under a parent. | – |
tickets_set_parent | Moves a ticket under a parent. | yes |
All tools except workflow_policy_get, workpackage_status and
tickets_children need Brave mode. See Session modes.
The preview
The preview shows the title of the action, the badge AI · with the tool name, and the steps under What will happen. Approve runs them, Cancel or closing the dialog rejects the call. Without the dialog, for example in a headless run, nothing is written.
For workpackage_open_pr the preview contains the title, the description and
the reviewer selection. Your changes there are used for the pull request.
For workflow_policy_set the preview shows the new policy as text. Show file
contents shows workflow.toml before and after the change.
Reviewer names that the AI passes are matched against the people the pull request provider offers: by login or ID first, then by a unique part of the name. An unknown or ambiguous name is refused with a list of the known reviewers. The AI then asks you instead of picking someone.
Setting up the policy from the chat
Describe to AI… in the Workflow Policy dialog starts this conversation. The
AI asks for what it cannot know, for example the statuses of your ticket
workflow, and checks them on a real ticket with tickets_transitions. It
writes the policy with workflow_policy_set, creates a decision entry in the
projectdocs from the returned text and links it in the policy. Later changes
are appended to that entry.