Skip to content

SOP-009: Artifact Publish & Index ​

Version: 1.0 · Owner: Platform / SRE · Gate: G1 (publish)

Purpose & scope ​

Publish Accepted ADRs and Approved specs from Git to human discovery surfaces and AI agent indexes.

Roles ​

RoleRACI
Platform / SRER/A — pipeline health
ARCHC — content validation
AIPOC — index/RAG config

Prerequisites ​

  • Merge to main of files under docs/adr/ or specs/
  • CI publish workflow enabled

Procedure ​

  1. Validate on merge — CI checks: ADR front matter valid, status allowed, links resolve, spec x-status: approved for indexing.
  2. Build static site — MkDocs/Docusaurus → S3 bucket artifacts.{org}.com.
  3. Invalidate CDN — CloudFront invalidation for /adr/* and /specs/*.
  4. Sync Backstage — TechDocs plugin pulls from Git or S3; catalog entity links updated.
  5. Re-index — OpenSearch Ingestion (or Kendra) ingests Accepted ADRs + Approved specs only.
  6. MCP refresh — ADR/spec MCP server cache invalidated or webhook triggered.
  7. Verify — Smoke test: MCP search_adr returns new doc within 15 min.

Definition of Done ​

  • [ ] Static site updated
  • [ ] Backstage shows new revision
  • [ ] Search index contains document (if status qualifies)
  • [ ] MCP returns document in smoke test

SLA ​

  • End-to-end publish: ≤ 15 minutes from merge

Failure handling ​

FailureAction
Index lagPage platform on-call; devs use Git path as fallback
Broken links in ADRCI should block merge; if prod index stale, hotfix PR

Reference template only — no production code. Adapt in your own project.