Roadmap and FAQ
Roadmap
| Milestone | Status |
|---|---|
| Offline analyzer, multi-document, source maps, 19 rules, profiles | Done |
| Plan summary without state | Done |
| MD + HTML + JSON reports and self-audit | Done |
| Read-only MCP server | Done (scripted client only) |
| Portable skills / agents | Done as files; native activation untested |
| Official YAML AST parser and schema validation | Planned |
| OPA/Conftest policy pack | Planned |
| Isolated Helm/Kustomize rendering | Planned |
| Read-only live drift and RBAC adapter | Planned |
| Canary/health/rollback evidence from external systems | Planned |
| Signed provenance | Planned |
| Dashboard and self-hosted RBAC server | Not started |
No dates are promised.
FAQ
Does it deploy anything? No. No command can apply, deploy, destroy or roll back, and the audit scans for that.
Why is there no "READY" status? Static checks cannot prove a release is safe. The only honest outcomes are "blocked" and "a human must decide".
Does it need a Kubernetes cluster or credentials? No. It reads files you give it.
Does it call an AI model or send data anywhere? No. No API key, no telemetry, no network.
Can it read Helm charts or Kustomize overlays? Not directly. Render them to plain YAML with a separately authorised tool, then analyze the output.
Why did it reject my YAML? It supports a strict subset and fails loudly instead of guessing. The error names the line and the construct.
Is --profile production stricter? Yes: MEDIUM findings block as well as HIGH.
Why does a clean manifest still show HUMAN_REVIEW_REQUIRED? By design. It also stays BLOCKED if rollback evidence is missing.
How accurate are the rules? On the 50-case corpus, exact. That corpus is the author's; expect misses and false positives on real manifests and report them.
Is a SHA-256 report digest a signature? No. It detects edits against a trusted copy only.
Is it a replacement for OPA/Gatekeeper/Kyverno? No. Those enforce at admission time. DeploySutra reviews before release and explains the change.
Which IDEs work? Skills are copied for Claude Code, Cursor, Codex, Gemini CLI, GitHub Copilot and OpenCode. Whether each host loads them was not tested. The CLI works everywhere Node 20+ runs.
Which platforms? CI runs Ubuntu, Windows and macOS on Node 20, 22 and 24.
How do I cite it? See CITATION.cff.