Skip to main content

CI/CD & workflow integration

Plug cryptarium into the pipelines you already run. It is a scan step that emits machine and human artifacts — SARIF, CBOM, JSON, Markdown — that CI gates, GitOps promote paths, and coding agents can consume. Why this matters: Post-quantum migration needs a repeatable inventory in the same place you already block bad merges and open fix work. Cryptarium does not replace your orchestrator; it feeds it.

Which artifact for which consumer

Pin CLI and Action refs to a tagged release (examples use v0.3.0).

Traditional CI and GitOps

GitHub Actions PR gate

Full job: gate on severity and upload SARIF to Security → Code scanning.
Start with critical, then tighten to high once the backlog is under control. Inputs and upload behavior: GitHub Action, Uploading SARIF, CI with —fail-on.

GitOps / scheduled inventory

Scan the application or infra repo that Argo CD, Flux, or similar syncs from — not the live cluster. Cryptarium inventories source, deps, certs, and config in that tree.
Add --fail-on critical (or wire the Action) on the promote path when you are ready to block syncs that still ship broken primitives.

Generic CI / GitLab

There is no first-party GitLab or Jenkins Action. The CLI is the integration surface — go install a tag or download a release binary (rule packs are embedded).
Exit codes: 0 success, 1 threshold met, 2 scan error. Artifacts are written before a threshold failure. See Flags & exit codes.

Agentic workflows

Cryptarium does not ship an MCP server or agent runtime. Agents and SecOps bots consume files the CLI already writes. Optional AI triage (Phase 4 roadmap) will annotate only — it must never create, delete, or reclassify findings, and the CBOM must stay identical with enrichment on or off.

Pipeline produces agent inputs

Emit stable JSON and Markdown in CI (alongside SARIF for humans and gates):
Upload out/ as a job artifact so a follow-on agent job, local Cursor/Claude session, or SecOps bot can fetch the same bytes every run. Prefer --deterministic whenever IDs or diffs feed automation.

Cursor / Claude Code loop

After a local or CI scan:
  1. Point the agent at out/CRYPTO-REPORT.md or out/cryptarium.json (default JSON name when --output is a directory).
  2. Use a fixed prompt that forbids inventing crypto claims:
  1. Human-review any draft PR before merge — the agent drafts; the inventory remains the source of truth.

SecOps bot pattern

On SARIF or JSON (webhook, cron, or Actions workflow_run):
  • Open or update issues for critical / high only.
  • Attach a short Markdown excerpt (location, class, recommended target, confidence).
  • Dedupe on stable finding IDs from deterministic scans.
Do not let a bot reclassify or drop findings; that stays in cryptarium’s cited classifier.
  1. Inventory only--fail-on none; retain SARIF + CBOM + Markdown.
  2. Gate critical--fail-on critical on default branch or promote.
  3. Tighten — move to high when the backlog is manageable.
  4. Optional agents — same artifacts into triage / draft-PR loops under human review.

What not to expect

  • No published Docker image for the CLI (container scanning is roadmap).
  • --policy is accepted but unused today (stub).
  • No first-party GitLab or Jenkins plugin — use the CLI.
  • The GitHub Action inputs today: path, fail-on, format (markdown|json|cbom), output, upload-sarif — not multi-format or CBOM upload as first-class inputs.
  • Static analysis cannot see runtime algorithm selection or wire negotiation. See Limitations.