Troubleshooting
The panel says the assessment won’t run
Provider not selected, or no key. ⚙ → Apps → Premortem settings → AI model. Choose a provider, paste a key, Save, then Test connection.
Test connection is the fast way to tell a bad key from everything else: it makes one real call and reports exactly what the provider said.
“This site has used its monthly limit”
An administrator capped how many AI runs the site may make this calendar month, and the cap is reached.
An admin can raise or clear it under Monthly limit, and the reading is shown by Check usage this month. Otherwise the allowance resets on the 1st, UTC.
The cap applies to administrators too, so seeing this as an admin is normal — raise the number and carry on.
“This action requires Jira administrator permissions”
Settings are admin-only, including reading them. Anyone who can see an issue can run assessments; only an admin configures the site.
Assessments are generic and could apply to any project
Working as designed, and fixable. In order of payoff:
- Point it at your architecture page — one Confluence link. Mitigations start naming your actual components instead of giving textbook advice
- Import your incidents — risks start citing what happened to you, by issue key
- Turn on comments under Context depth — decisions and caveats frequently exist nowhere else. It costs tokens and it is usually worth it
- Write a standing instruction for what is permanently true about your context
An assessment on a one-line issue description will be thin no matter what. The model cannot know what was never written down.
Nothing appears in the Code risks tab
Work through it in this order:
| Check | |
|---|---|
| Does the branch or PR title contain the issue key? | No key, no assessment. BT-12-fix-webhook or [BT-12] ... |
| Did the Action run at all? | Look at the workflow run in GitHub |
| Did it log a 401? | PREMORTEM_CI_SECRET is wrong or missing |
| Did it say “no CI signing secret”? | The site never fetched the webhook URL. ⚙ → Apps → Premortem settings → CI integration |
| Did it succeed and still nothing? | Correct. CI only stores the diff — open the issue and click Assess code risks |
The pipeline broke right after rotating a secret
Expected. Rotating the CI secret or regenerating the webhook URL invalidates the old one immediately. Update the GitHub secret — one organization secret update, not one edit per repository.
Imported incidents are being ignored
The settings page shows a count of incidents embedded with a different provider than the one now configured. Those are skipped, because vectors from different models cannot be meaningfully compared.
Re-import the project.
The register has rows for issues that no longer exist
⚙ → Apps → Premortem settings → Register maintenance. It also runs on a schedule; the button is for when you want it correct now.
Committing the same issue twice duplicated the risks
It should not — rows and the description section are replaced, not appended. If you are seeing genuine duplicates, that is a bug worth reporting with the issue key and the register page.
Two Premortem panels on the issue
One is a development build. Only one should be installed on a production site; if you see a panel badged DEVELOPMENT, an admin installed a dev version of the app on this site.