Skip to main content

Issue trigger

An issue-triggered skill runs when a work item in your issue tracker is of a given type and in a given status. This is how Codee does development work: it picks the issue up, does the work, and moves the issue to the next status.

---
name: task-developer
description: Implement a standalone Issue Tracker task and submit a pull request.
disable-model-invocation: true
x-codee-trigger: issue
x-codee-issue-status: ['AI Ready for development', 'AI In Progress']
x-codee-issue-type: task
argument-hint: <TASK_ID>
---

Setup​

1. Connect a tasks provider​

Issue triggers are the only trigger that needs a tasks provider. Set up Jira or Azure DevOps under Settings → Tasks provider and use Verify connection to check it. See Configuration.

Nothing is polled until the provider is configured.

2. Map the work items​

Codee has its own work items, and each one is mapped to the work item types your provider actually uses. story and task always exist:

Codee work itemJiraAzure DevOps
storyStoryUser Story
taskTaskTask

Add a row under Settings → Tasks provider → Work items for anything else you want skills for. See Configuration.

3. Create the statuses​

Create the statuses your skills will trigger on, and the transitions between them, in your provider's workflow.

The two tutorials walk through this for each provider:

4. Declare the trigger​

x-codee-trigger: issue
x-codee-issue-type: task
x-codee-issue-status: ['AI Ready for development', 'AI In Progress']

x-codee-issue-type takes one Codee work item, and it has to be one of the work items configured in Settings. A skill for a bug row you added yourself uses x-codee-issue-type: bug.

x-codee-issue-status takes one status or a list. Names are matched case-insensitively, but otherwise they have to match the provider exactly — AI Ready for CR and AI ready for code review are different statuses.

5. Say where the issue goes next​

Codee does not move the issue for you. The skill body tells the agent to do it:

Move the task to `AI In Progress` before doing any work.
Move it to `AI Ready for CR` once the change is complete and its pull request is open.

Moving the issue out of the triggering status is what stops it being picked up again on the next poll. A skill that triggers on AI Ready for development and leaves the issue there runs again every minute.

The Workflow page draws the statuses and the skills that connect them, which is the fastest way to see where an issue can get stuck.

How the match works​

On each tick Codee asks the provider for issues in every status any issue skill triggers on, narrowed by the mapped work item types and by Custom JQL / WIQL if you set one. For each issue it finds the first skill whose issue type and status both match, and runs it as:

/task-developer PROJ-4192

The skill name in that command is the folder name under .claude/skills/, so keep the folder name and the name field the same.

One issue type and status pair should be claimed by one skill. If two skills claim the same pair, the first one found wins and the other never runs.

Things worth knowing​

  • Children of a mapped work item are skipped. An issue whose parent is of a type mapped to a Codee work item is driven by the parent's own run rather than picked up on its own. With the default mapping that covers a Task under a Story and a Task under a Task. Only the types on a row count: a work item written as a JQL or WIQL condition takes no children out of the queue, since nothing in a parent says whether that condition claims it.
  • A task already running is not started twice. An issue with an agent in flight is skipped on later ticks until that agent finishes.
  • Failures are retried. If the agent fails, the issue keeps its status, so the next poll picks it up again. Both attempts are on the Runs page.
  • Over the parallel cap, issues wait. Anything beyond Max parallel tasks keeps its status and is picked up on a later tick.
  • Every run is a fresh session. Codee never resumes a previous one, so a skill that picks work back up has to find its own state: the issue's comments, the branch, the open pull request.