Sentinel Vault: Runs on Atlassian AI apps
Description and use
Sentinel Vault uses Atlassian Forge LLMs API to review Confluence content without external keys or egress. The production build meets Runs on Atlassian criteria, but the development build can egress data via a web trigger. AI review is off by default and requires admin activation, with token usage billed to the vendor’s Forge account.
Problem
Confluence admins must decide whether to enable AI review in Sentinel Vault, but the marketing claim of no external data egress and no API key may be misleading without code verification.
Outcome
The app satisfies the Runs on Atlassian badge requirements in production, but the development build can egress data via a web trigger. AI review is disabled by default and requires admin activation; token usage is billed to the vendor’s Forge account.
Workflow
- 1. Review the app’s manifest.yml for external permissions and modules.
- 2. Run
forge eligibilityon production and development builds to check Runs on Atlassian eligibility. - 3. Inspect the deployment script
scripts/deploy-prod.shto confirm removal of dev modules. - 4. Examine default configuration in logic.js to verify AI is off by default.
- 5. Verify that only admins can trigger AI reviews and that token usage is billed to the vendor’s Forge account.
Evidence
Sentinel Vault is one of the Runs on Atlassian apps that review Confluence content with AI, through the Forge LLMs API, with no API key to paste and no external data egress declared.
$ forge eligibility -e production The version of your app [6.5.0] that's deployed to [production] is eligible for the Runs on Atlassian program.
$ forge eligibility -e development The version of your app [8.98.0] that's deployed to [development] is not eligible for the Runs on Atlassian program. - App is using a webtrigger module that can egress data
The default configuration in logic.js sets the validation engine's master switch to enabled: false. AI has its own switch, ai: { enabled: false, ... }
Token costs bill to the app vendor's Forge bill, so we enforce a Haiku-only policy (see isForgeLlmModelAllowed) at every layer.
- Record date: 2026-10-02T05:55:42.078Z