Agent-Ready Check for Jira

Documentation

Agent-Ready Check scores each Jira work item from 0 to 100 on whether an AI coding agent can start on it, and lists what to fix. This guide covers setup and everyday use.

1. Install

A Jira admin installs the app from the Atlassian Marketplace and approves its permissions. There is nothing else to set up: every project starts with the default settings below, and work items are scored as they are created, edited or opened.

2. See a work item's score

Open any work item. The Agent-ready badge shows the verdict and score, for example "Ready 92" or "Not ready 40". Open the Agent-Ready Check panel to see each failed check with a one-line fix. Rescore checks the work item again.

A work item is ready when its score is at or above the project's threshold (75 by default) and no blocker check fails. Epics and initiatives are not checked.

3. The checks (ruleset 1.0.0)

Check Name What passes Default
AR-01 Testable outcomes listed At least 2 checkable outcomes under a heading such as "Acceptance criteria", or one "Then" per Given/When/Then Blocker, weight 15
AR-02 Outcomes are where agents look Outcomes are in the description or an "Acceptance criteria" field, not only in comments, attachments or other fields Warning, 5
AR-03 Finish line defined A "Definition of done" section Warning, 8
AR-04 Codebase is pinned A component mapped to a repository, a repo: label, or the repository URL in the description Blocker, 12
AR-05 Risk level declared Exactly one risk label (risk-low, risk-medium or risk-high) or a Risk field Blocker, 8
AR-06 How to verify A test plan, a test command, or tests to add or update Warning, 8
AR-07 Precise verbs No fuzzy words such as "improve" or "clean up" Warning, 6
AR-08 No hidden context No references such as "as discussed" without a link Warning, 6
AR-09 Enough detail At least 200 characters of description; bugs need steps to reproduce, expected and actual Blocker, 8
AR-10 Sized Story points of 5 or less, or a size of M or less Warning, 4
AR-11 Sensitive areas flagged Work touching areas such as auth, payments or personal data has risk set to high Blocker, 10
AR-12 UI work has a design UI work links a design or attaches a mockup Warning, 5
AR-13 Dependencies resolved No open "is blocked by" links Warning, 3
AR-14 One thing at a time The summary describes one change, under 120 characters Warning, 2

The score is the share of applicable weight that passes. Checks that don't apply to a work item are left out.

4. Project settings

Project admins and Jira admins can change the settings in Project settings → Apps → Agent-Ready Check:

  • The ready threshold (0 to 100)
  • Each check: on or off, blocker or warning, and its weight
  • The override label (default agent-ready-override)
  • A map from component names to repositories, and extra repository names that count as pinned

Save applies the settings to that project. Reset to defaults goes back to the defaults.

After a save or reset, existing scores in the project show "rules changed". The app re-scores each work item when someone opens or edits it, and a background job re-scores the project's scored work items, 20 at a time, 30 seconds apart.

5. Block unready work with the validator

A Jira admin can add the validator to a workflow transition, such as To Do → In Progress:

  1. Edit the workflow and select the transition.
  2. Add a validator and choose Work item is agent-ready.
  3. Pick a mode: Fully agent-ready (the default) blocks the transition unless the work item is ready. No failed blockers blocks it only when a blocker check fails.

When it blocks, the message shows the score, the threshold and up to three failed checks. The validator always scores the work item as it is now, with the current settings.

  • Skip it: add the override label to the work item.
  • If the app fails (an error, more than 10 seconds, or a work item it can't read), the transition goes through. The validator is a helper, not a security or compliance control.

6. Search scores with JQL

The app saves each score on the work item, so you can search it:

  • ab500ReadyVerdict (ready, not_ready or not_applicable)
  • ab500ReadyScore, ab500ReadyBlockers
  • ab500ReadyFailedRules (for example AR-04)
  • ab500ReadyScoredAt, ab500ReadyRuleset, ab500ReadySchema

Example: project = ABC AND ab500ReadyVerdict = "not_ready" AND statusCategory != Done ORDER BY ab500ReadyScore ASC

Saved scores update when a work item is created or edited, when someone opens a "rules changed" work item, and after a settings change. There is no nightly sync, so a change that doesn't touch the work item itself, such as a linked blocker closing, shows in JQL after the next edit or view.

7. Rovo and MCP

In Rovo, chat with Agent-Ready Coach. Ask "Is this work item ready for an AI agent?", "Which work items in my project are not agent-ready?", or "Rewrite this work item so it is agent-ready". The tools are read-only: Rovo drafts a better description for you to paste, and never edits the work item.

A site admin can also let outside AI tools use the same three tools through Atlassian's MCP service. This is off unless an admin turns it on. See section 6 of the Privacy Policy.

8. Help