In March 2026, we made an unconventional infrastructure decision for the Wikantik publishing stack. Rather than adopting an off-the-shelf headless CMS like Ghost, Strapi, or WordPress, we inherited and forked Apache Roller 6.1.5—an enterprise multi-tenant blogging platform originally created in 2002. At first glance, Roller looked like an archaeological relic: it was encumbered by twenty-four years of accumulated Jakarta EE baggage, an Apache Struts routing layer, XML-RPC pingback handlers, legacy blogroll managers, six obsolete proprietary SQL database dialects, and an operational deployment model strictly tethered to external servlet containers.
Yet beneath two decades of sediment lay something increasingly rare in modern software: an exceptionally robust relational data model for multi-tenant weblogs, battle-tested permission hierarchies, and a proven, clean separation between editorial authoring and high-performance template rendering. What Roller lacked was not domain integrity; it lacked modern architectural boundaries, a streamlined runtime, and first-class support for autonomous artificial intelligence agents.
This inaugural post on the Pressroom development journal documents the complete trajectory of the platform so far. Across eight months of disciplined engineering—culminating in the recent 0.8.8 release—we executed a radical simplification, rebuilt the runtime on Java 25 LTS and Spring Boot, orchestrated a seamless production cutover, and pioneered an agent-first publishing architecture governed by the Model Context Protocol (MCP).
1. The Pruning Wave: "Done When There Is Nothing Left to Remove"
Modernization projects routinely fail because teams attempt to refactor legacy codebases while preserving every historical feature. Our governing philosophy from day one was Antoine de Saint-Exupéry's classic maxim: perfection is attained not when there is nothing more to add, but when there is nothing left to remove.
Before writing a single new feature, we executed a sweeping deletion campaign (Phase 0 and Approach B Simplification) that permanently eliminated tens of thousands of lines of dead weight:
- The Planet Feed Aggregator (~4,200 LOC, 6 DB Tables): Apache Roller historically bundled an RSS/Atom aggregator called Planet. In an era of specialized feed readers and vector indexes, embedding an entire feed crawler inside a blogging engine was an anti-pattern. We excised the entire
org.apache.roller.planet.*tree and dropped tablesrag_planet,rag_group,rag_subscription, andrag_entry. - Bookmarks and Blogrolls (~1,500 LOC, 2 DB Tables): The 2004-era blogroll management subsystem was deleted from controllers, JPA managers, Velocity models, and the database schema (dropping
roller_bookmarkandroller_bookmarkfolder). - XML-RPC Pings, Trackbacks, and External Ping Queues (~1,800 LOC, 3 DB Tables): In the early 2000s, weblogs notified search engines and other blogs of new articles via XML-RPC pings and trackback URLs. In modern production, trackbacks are exclusively an attack vector for comment spam and distributed denial-of-service. We eliminated all ping targets, queue processors, and dropped tables
roller_pingtarget,roller_autopings, androller_pingqueueentry. - Purging Six Proprietary SQL Dialects: Upstream Roller maintained divergent DDL and custom SQL mappings for Oracle, IBM DB2, Apache Derby, HSQLDB, Microsoft SQL Server, and MySQL. We standardized exclusively on PostgreSQL. Eliminating polyglot SQL dialects enabled us to leverage advanced PostgreSQL primitives, transactional DDL, and native
pg_dumppipelines without abstraction leakage. - Markdown-Only Content Pipeline: Upstream Roller supported multiple divergent formatting plugins (HTML, XHTML, Wiki, Markdown). We eliminated format fragmentation by standardizing strictly on Markdown powered by
commonmark-java. By running AST processing after shortcode expansion and prior to OWASP-compliant HTML sanitization, Pressroom established a single deterministic content pipeline.
2. Modernizing the Foundation: Spring Boot, Java 25 LTS, and Strict DI
With legacy fossils eliminated, the core infrastructure required modernization to eliminate servlet container complexity and operational fragility.
Eliminating the Static Service Locator
Upstream Roller relied heavily on WebloggerFactory, a global static singleton service locator that made unit testing brittle and introduced hidden temporal coupling during application startup. On August 22, 2026, we permanently deleted WebloggerFactory. In its place, we instituted clean constructor dependency injection using Spring Boot:
- Business beans and MVC controllers receive a
@Lazy Webloggerfacade directly through constructor injection. - The lazy boundary is enforced through architectural gates:
ContextRefreshDoesNotBootstrapTestandStaticServiceLocatorTestfail CI builds if any static business-tier accessor is reintroduced. - Domain entities under
com.wikantik.pressroom.pojosare pure data plus business invariants, importing zero business-tier managers or persistence context types.
The Java 25 LTS Commitment
Pressroom runs on Java 25 LTS. On September 15, 2026, the team established an explicit architectural invariant: the runtime remains pinned to Java 25 LTS until Java 29 LTS arrives in September 2027. We explicitly avoid intermediate six-month non-LTS releases. Seven synchronized pins across the codebase—the Maven enforcer range [25,26), the compiler release target, PMD static analysis rules, GitHub Actions workflows, and Dockerfile Eclipse Temurin base images—guarantee predictable, rock-solid stability.
Zero-Tolerance Credential Security
Legacy configurations frequently permitted plaintext passwords or reversible DES encryption schemes for administrative accounts. Pressroom completely eradicated plaintext and legacy hash algorithms. Password encryption is mandatory and irreversible: bcrypt is the default, with optional scrypt and argon2 backends. Database-backed authentication fails loudly at startup if invalid auth methods are configured, and no setting or environment variable can bypass cryptographic hashing.
Pressroom follows a strict fail-closed philosophy across all operational boundaries. If an invalid authentication scheme is passed, if a required database migration fails, or if an API token exceeds its cryptographic boundary, the runtime halts immediately rather than silently degrading into an insecure state.
3. The Agent-First Revolution: Native Model Context Protocol (MCP)
The defining architectural breakthrough of Pressroom is its elevation of autonomous AI agents to first-class citizens. Conventional content management systems are built under the assumption that a human sits behind a graphical browser clicking form buttons. In the Simple Agility ecosystem, autonomous agents continuously draft, edit, publish, audit, and analyze content across multiple technical publications.
To eliminate the brittle failure modes of web scraping and ad-hoc REST calls, Pressroom introduced two synchronized machine-to-machine interfaces:
- Scattered REST API (
/api/v1): A typed REST surface authenticated via cryptographic Bearer tokens (prs_...). Tokens support granular, least-privilege scoping:- Roles:
READ,POST, orADMIN. - Weblog Pinning: An API token can be strictly confined to a single weblog handle (e.g.,
investingorpressroom). Any attempt to read or mutate another weblog returns an opaque HTTP 404, preventing IDOR information disclosure.
- Roles:
- Native Model Context Protocol Gateway (
/api/mcp): Exposing twelve comprehensive operational domains directly to LLM runtimes over JSON-RPC 2.0:entries: Full lifecycle management (create, update, draft, publish, schedule, diff, and revisions).categoriesandmedia: Hierarchical media folder structures, asset localization, and taxonomy management.weblogsandpages: Multi-tenant provisioning, theme configuration, and custom page authoring.audits: Automated server-side SEO inspection, link integrity, and snippet validation.site_settings,users, andmaintenance: Comprehensive administrative authority.whoami: Runtime credential inspection and permission discovery.
4. The Production Cutover: Renaming the Stack Under Load (v0.5.0)
By late October 2026, the codebase had diverged so completely from upstream Apache Roller that maintaining the historical branding caused continuous cognitive dissonance. In release 0.5.0, we executed a total identity cutover.
Executed via automated scripts (cutover-pressroom.sh) and proven in rehearsal on cloned staging environments, the 0.5.0 cutover performed a complete, seamless rename across every layer of the production stack:
- Flyway Migration
V041: Atomically renamed all database tables fromroller_*topressroom_*and updated primary sequences. - PostgreSQL Roles and Datastore: Renamed the PostgreSQL user and database to
pressroom. - Container and Volume Topology: Re-keyed Docker Compose volumes, network boundaries, and container namespaces from
ghcr.io/jakefearsd/pressroom. - Environment Re-keying: Transformed all
ROLLER_*configuration keys toPRESSROOM_*without touching a single secret value.
The entire production migration completed with zero data loss, verified row count parity across all application and analytics tables, and clean rollback guarantees.
5. Modern Audience, Back-Office Decoupling, and "Pictures by Default" (v0.8.x)
Following the 0.5.0 cutover, development focused on audience engagement, performance optimization, and asset resilience across the 0.8.x series:
Google SSO Reader Accounts and Spam-Proof Comments
Anonymous blog comments are an intractable magnet for automated spam. Pressroom solved this by integrating lightweight OpenID Connect (OIDC) reader sign-in via Google SSO. Readers sign in with their existing Google account; their identity is verified via OAuth 2.0; and they can participate in high-quality discussions without passwords or captchas. Crucially, reader accounts are strictly isolated: they possess zero access to admin routes, authoring UIs, or API endpoints.
Back-Office Decoupling (v0.8.5)
To provide modern newsletter delivery and privacy-respecting analytics without bloating the core Java runtime, Pressroom integrates with bundled Listmonk and Umami instances. In release 0.8.5, we decoupled these back-office services from Pressroom's primary deployment bundle. The web application interacts with them across defined internal API boundaries, allowing independent versioning, isolated database lifecycles, and horizontal scalability.
"Pictures by Default" and Fail-Soft Media Processing (v0.8.7)
In analytical and travel publications, visual fidelity directly influences reader retention. Prior to version 0.8.7, publishing agents were required to supply explicit featuredImageUrl and featuredImageAlt parameters alongside body HTML. If an agent omitted these parameters, the post rendered with a fallback colored initials tile in front-page discovery feeds.
Pressroom 0.8.7 established an intelligent server-side media pipeline:
- Automatic Lead Image Promotion: When an entry is saved without an explicit featured image, Pressroom parses the content AST (
LeadImages.mediaIds(text)), promoting the first embedded image tofeaturedImageIdautomatically. - Best-Effort External Ingestion: External image URLs embedded in posts are localized into the weblog's first-party media library, generating responsive WebP renditions (480w, 960w, 1376w) and BlurHash previews. In accordance with fail-soft design principles, if an external host fails or rate-limits, the asset remains linked without aborting the write transaction.
- Hero De-Duplication: Modern themes automatically compare body figures against the featured hero image, suppressing redundant duplicate banners above the post headline.
Discovery Aggregation and Traffic Intelligence (v0.8.8)
The most recent release, 0.8.8, introduced high-performance front-page discovery aggregation. Instead of executing N+1 queries across multi-tenant blogs, Pressroom counts posts and queries recent cross-weblog activity in single-pass SQL statements. Furthermore, autonomous agents gained direct traffic telemetry via GET /v1/weblogs/{handle}/traffic and MCP tool traffic get, allowing autonomous publishing pipelines to monitor audience engagement in real time.
The Road Ahead for Pressroom
Pressroom began as an ambitious modernization experiment: proving that a venerable twenty-year-old open-source application could be pruned of historical debt, stabilized on modern Java LTS primitives, and transformed into an agent-grade publishing engine. Today, it powers the entire Simple Agility publishing network with sub-millisecond query responses, automated SEO validation, and seamless human-agent pair publishing.
As we look forward to the 0.9.x milestones, our engineering roadmap remains focused on deep agent collaboration: expanding automated SEO remediation, providing asynchronous batch archive ingestion, and refining theme portability across multi-channel syndication. For live architectural documentation, operational runbooks, and how Pressroom fits alongside Wikantik and JakeMon in the Simple Agility stack, visit the Pressroom Hub on the wiki.
Welcome to Wikantik Blog, Blog. This publication is where every release, architectural decision record, and production lesson of Pressroom will be documented in open, verifiable detail.
Comments