A Jira Cloud app on the Atlassian Marketplace
Your AI agents and workflows, as buttons on the ticket.
Fix the bug, rewrite the description, test the preview, deploy. Each button runs Claude Code, Codex or any GitHub Actions workflow, only for the Jira groups you pick, and the result offers the next one.
SHOP-142
Checkout fails when a saved card is used
Activity
- Paying with a saved card returns a 500.
Next steps
Elif Demir · Developers
Actions
Start with Send to agent. When a button is locked, switch roles.
From the ticket to your agents and back
Your agents run in your GitHub Actions with your own API keys. Cuelight itself uses no AI: it checks who pressed, passes the fields you choose to your workflow, and the workflow decides what the agent sees.
Press a button
Someone presses Write description with AI, Send to agent or Deploy on the ticket. Cuelight checks their Jira groups and starts your workflow with the ticket's key and any fields you choose.
Your agent works
Your AI agent or job runs in your GitHub Actions. It rewrites the description, opens a pull request or tests the preview, and reports back to the ticket.
Next step
The result picks the next buttons, like Test the preview, Promote to production or Run agent again, each for the Jira groups you pick.
See it in 30 seconds
The same loop works without an agent. A tester presses Deploy to staging on a work item. Cuelight checks their Jira group, runs the workflow, and the job reports back with the next buttons.
One step in your workflow
Add it at the end of the job. The result and the next buttons land on the work item.
- name: Hand back the next step
if: always()
uses: huseyinbuyukdere/cuelight-report@v1
with:
status: ${{ job.status }}
message: Deployed to ${{ inputs.environment }}
next: rollback-staging, promote-productionDeploy to staging
Deployed to staging
2m 14s, by Maya Chen
Suggested by deploy.yml
Checked before it runs, written down after
- Jira groups are checked on the server when someone presses, not only when the panel loads.
- Every run is recorded: who pressed it, what they entered and how it ended.
- Each run gets its own signed token that expires after 24 hours.
Only members of Release Managers can run this.
Activity
- ▶ Deploy to staging started by Maya Chen. Inputs: environment=staging, reason="QA pass 2"
- ✓ Deploy to staging succeeded in 2m 14s.
What it replaces
Compare with Jira Automation and othersAn agent that stops at the pull request
Someone still has to deploy a preview, approve it and tell the ticket.
A label that starts a workflow
Anyone who can edit the ticket can add it, and nothing comes back.
An Automation web request
It can start a workflow, but the result and the next step stay in GitHub.
What teams use it for
All recipesRun Claude Code or Codex from Jira
Send a ticket to Claude Code or Codex. The pull request comes back with the next buttons.
Write the Jira description with AI
Turn a one-line bug report into steps and acceptance criteria with your AI.
Test the preview with AI
Your AI writes and runs tests on the preview. Only a pass offers Promote.
Approve after an agent's pull request
An AI agent opens the pull request. A person approves the deploy from Jira.
Deploy, promote and roll back
Deploy to staging, promote to production and roll back, each with one press.
Keep your label automations
Put the label-based Jira Automation rules you already have behind a button.
Set up your buttons without JSON
Start from a template, answer a few questions and try the buttons on a sample work item. Or describe your buttons to your own AI chat and paste its answer in to check it. Then copy the result into Jira.
- Write description with AIRuns ai-description.ymlDevelopers and QA, on every work item
- Send to agentRuns agent.ymlDevelopers only
- Test the preview with AIRuns ai-tests.ymlOnly when a preview deploy offers it
- Promote to productionRuns deploy.ymlRelease managers, after the tests pass
Free for up to 10 users
Larger teams get a 30-day free trial, then pay per user. See pricing