AI Delivery Intelligence v0.1.0 is the first public release of an evidence-grounded decision-support product for continuous delivery health and management attention.
Highlights
- Source-agnostic normalized delivery model shared by Demo, Kaiten, and Jira adapters
- Persisted snapshots and deterministic change detection across reviews
- WIP, aging, blocker, dependency, ownership, overdue-work, and commitment-risk signals
- Replaceable Markdown policies with inspectable BM25 retrieval and concrete citations
- Bounded read-only Responses API tool loop with strict structured output
- Evidence and policy reference validation, uncertainty handling, and safe replay fallback
- Credential-free Northstar Platform T1 → T2 demo with run history
- Prompt-injection safeguards for untrusted tracker content
Verification
- 35 backend tests, Ruff, and strict mypy
- Frontend component test, typecheck, and production build
- Playwright E2E over the core demo journey
- Docker Compose startup plus API health smoke test
- 25/25 versioned evaluation scenarios, including cross-adapter equivalence, policy retrieval, malformed output, unsupported references, and prompt injection
Product boundary
ADI reports continuous delivery health and management attention. It does not inspect CI, pull requests, release branches, migrations, deployments, or make Go/No-Go or release-readiness decisions.
Limitations
One project or board per run; no cross-project portfolio intelligence; no Jira–Kaiten aggregation; no write-back, notifications, or automatic escalation; source capabilities vary; Jira workflows and relation names may require mapping; policy quality affects recommendations; public data is synthetic; no production-adoption or business-impact claim.