Sunday, 8 February 2026

I built a “Steam Workshop” for system architecture + product roadmaps + org blueprints (runs in your browser)"

Have you ever wanted to install the architecture of a product the way you install an app?

Not just a diagram, but the whole system architecture blueprint:

  • services and dependencies
  • teams and ownership
  • goals → initiatives → work packages
  • a 3-year planning horizon for product managers
  • plus a “prompt pack” so you can remix it

That’s what I’ve been building as a hobby project: SMT (Software Management & Planning Tool).



And this week I shipped something I’m genuinely excited about:

The Community Blueprints Marketplace (with social features)

You can now publish blueprint packages publicly, and the community can:

  • browse and install them
  • star them
  • comment / discuss them
  • see what’s trending (social proof + discovery loops)

Think: Figma Community / Steam Workshop, but for product architecture + product/team organization.


What’s a “Blueprint” in SMT?

A blueprint is a portable package (JSON) that contains:

  • manifest (title, summary, tags, trust label, etc.)
  • prompt pack (seed + variants like MVP/Scale)
  • full system snapshot (teams, services, goals, initiatives, work packages)

The goal is learning through interaction:

  • install a blueprint
  • explore its org + architecture + roadmap in the app
  • remix it into your constraints
  • publish your remix back to the marketplace

Why I’m doing this

Most “reference architectures” online are:

  • static
  • divorced from org/team realities
  • not easily remixable
  • missing the roadmap/execution story

SMT tries to make “how a product might actually run” tangible:

  • architecture + org design + planning are all connected
  • you can poke at it, not just read it

SMT makes it possible for you to inspect any type of tech platform. Think LeetCode interview preps but for system design, architecture, team topologies, product roadmaps and software delivery planning.


Local-first by default (privacy + zero friction)

SMT is intentionally local-first:

  • it runs as a static app in the browser
  • your systems stay in your browser unless you explicitly publish a blueprint package

The cloud marketplace is optional and only powers:

  • public publishing
  • discovery
  • stars/comments

No “workspace sync” SaaS lock-in.


How the social marketplace works (simple + cost-free)

To keep this sustainable on the free tier, the backend is:

  • Cloudflare Workers + D1 (SQLite)
  • token-normalized search (no paid search / no vector DB)
  • GitHub OAuth for identity (scope: read:user only)

Important security bit:

  • public publishing does secret scanning (manifest + full system payload) and blocks likely API keys/tokens.

Try it (and please break it, it's a WIP hobby project)

1) Open the app

2) Explore the “Community Blueprints” view

  • browse the curated catalog
  • click Preview on anything that looks interesting
  • install an Available blueprint and inspect it across:
    • System Architecture Overviews
    • Org Design
    • Roadmap & Backlog Management
    • Year Plan / Detailed Planning

3) Publish + socials

  • in the publish flow, use Publish Publicly
  • then open Preview on your published blueprint:
    • Star/Unstar it
    • Post a comment
    • sanity check that it’s discoverable via search

If anything fails, I want to know. Use the Feedback feature to log issues.


What I’d love feedback on (high signal)

  1. Does the blueprint concept actually help you understand a product faster?
  2. Are the “prompt packs” useful, or just noise?
  3. What should “trending” mean here: stars, downloads, recency, or something else?
  4. What social features would make this fun without turning it into a moderation nightmare?

If you want to contribute

This is open source (CC0) and I’m happy to collaborate.


Roadmap ideas (if the community likes this)

  • Remix lineage: “Forked from…” + remixes graph
  • Lightweight contributor reputation (badges, trust tiers)
  • Reporting/flagging + moderation queue
  • Curated collections (“Backends 101”, “B2B SaaS starters”)

If any part of this sparks your curiosity, I’d love for you to try it for 5 minutes and tell me what confused you, what felt magical, and what felt pointless.

Drop a comment here, or open an issue on GitHub.

Saturday, 7 February 2026

A Day Building new features on SMT using Codex App with Codex 5.3

AI Build Journal · February 7, 2026 - written by Codex to me...OpenAI released codex app for mac this week, so I decided to have a go, and boy - am I blown away!!! In just one day, Codex helped me clear much of my SMT backlog, after a month's break from my AI-coding frenzy from Dec'25.

A Day Building on Codex App with Codex 5.3

This was not “prompt in, code out.” This was a full-day product session: strategy debates, UX corrections, contract audits, feature pivots, test hardening, documentation, and ship.

I started the day with one objective: execute the next phase plan for SMT Platform without losing quality. By the end of the day, we had shipped one of the most ambitious increments in the project so far: Goal Inspections + Community Blueprints Exchange, including end-to-end contribution flow, install lifecycle logic, catalog operations, and test coverage.


What We Shipped in One Day

  • Goal lifecycle + inspections system: owner status, weekly comments, PTG tracking, stale/mismatch detection, leadership report table.
  • Year Plan CSV/XLSX export: production export flow in toolbar with tested serialization and schema-aware payload handling.
  • Community Blueprints Exchange: Top-100 curated catalog, preview modal, search/filter, publish flow, package validation, and install lifecycle UX.
  • Launch package generation upgrade: moved to domain-authored-curated-v2 for the launch-25 package set.
  • Hardening + compliance: contract remediation pass, UX consistency fixes, event rebinding bug fix, and regression-proof e2e updates.

The Metrics That Matter

Metric Result
Session duration11h 56m 43s (10:08:30 → 22:05:13)
Timestamped worklog checkpoints        117
Commits shipped4 (2671b68, a502106, c9d8d43, 97bec52)
Code delta (same-day commits)+129,166 / -7,770 (net +121,396)
Unique files touched77 (113 file-change events across commits)
New files created27
Unit test progression90 → 117 tests (+30%)
E2E test progression51 → 58 tests across 8 → 9 specs
Community blueprint footprintTop-100 catalog + Launch-25 curated packages

Note: the large insertion volume includes generated blueprint catalog/package artifacts in addition to application code.

How This Compared to “Typical Solo Dev Pace”

A conservative estimate for this scope with one human engineer is 2–3 weeks: feature architecture, UI wiring, persistence, migration work, docs, and full regression coverage. Here, the value of Codex 5.3 was not just speed in typing code. The leverage came from:

  • Staying in implementation mode continuously while preserving test discipline.
  • Switching quickly between product decisions, coding, debugging, and documentation.
  • Keeping a verifiable trail (/docs/worklogjournal.md) so context did not get lost.

This Was Collaboration, Not Task Dispatch

The most important part of the day was the interaction pattern. We did not run a one-way backlog. We debated quality and credibility:

  • You challenged weak UX states (Install should be locked when unavailable), and we corrected behavior at both tile and preview levels.
  • You challenged data realism for “inspired-by” systems, and we replaced simplistic seed generation with richer domain-authored package generation.
  • You enforced coding contracts, and we ran an explicit compliance audit plus remediation pass before final push.
  • You required proof, not promises, so every major change ended with lint/unit/e2e verification.
The real unlock is not “AI writes code faster.” It is “human judgment + AI execution + strict verification” as one continuous loop.

Lessons Learned

  1. Contracts first, always: when contract rules are explicit, quality issues become detectable and fixable quickly.
  2. Feature credibility beats feature count: shipping a marketplace means realism, not placeholder parity.
  3. Tests are collaboration memory: every bug found late became a permanent test so the fix does not regress.
  4. Worklogs scale agentic development: detailed timestamped logs made long-session continuity possible.

What’s Next

The obvious next move is to raise the “real-world blueprint” bar further: richer domain fidelity, stronger package QA gates, and a true contribution-driven exchange loop where users generate, validate, publish, and learn from each other’s systems.

Built on SMT Platform using Codex 5.3 · evidence from /docs/worklogjournal.md and same-day git history.