Cloudy with a chance of tech debt — a free, browser-based org assessment. Nothing uploads.
You can smell a struggling org before you can prove it.
You open one you’ve never seen, and within minutes your gut has a verdict. Too many triggers. A permission set nobody remembers assigning. A flow that’s been “temporary” since 2021. You know it’s rough.
But “my gut says it’s rough” isn’t an assessment. And turning that instinct into something a customer can act on — a scored, defensible, this-is-what-to-fix-and-why report — used to cost me two days of exporting spreadsheets and squinting at metadata to say what I already knew in two minutes.
That gap — between what an architect senses and what they can defend — is the thing I got tired of.
The problem with scans
There are good tools in this space, and I use them. sfdx-hardis collects and deploys. Org Check inventories your technical debt. Code Analyzer lints. They’re excellent at “what exists?”
But that’s not what the customer is really asking. They’re asking “was this built right?” — and that’s a judgment, not a scan. A class with SOQL in a loop isn’t just “present”; it’s a decision that will hurt at volume. A custom object that mirrors standard Contact isn’t “an object”; it’s a fork in the road someone took, maybe for a good reason, maybe not.
Scanners tell you what’s there. They don’t tell you whether it holds up.
Meet orgspark
For years I’ve built the layers to close that gap on real projects — the checks, the scoring, the reuse-vs-rebuild calls, the discipline of grounding a verdict so it survives the room. orgspark is those layers, finally put together.
A Salesforce org assessment that runs entirely in your browser.
It scores your org against Salesforce’s own Well-Architected guidance, lists every component with its own verdict — ✅ OK, 🟡 Watch, 🔴 Fix, the healthy ones included — and cites where each verdict comes from. About 80 checks across security, code, automation, data model, agents, adoption and governance.
What it won’t do
Here’s the part people don’t believe at first: nothing uploads. There are no servers. orgspark is a static web page — it parses your metadata, runs the checks, scores it and renders the report entirely inside your browser tab. The only network call it’s even allowed to make is to your own org (read-only) and, if you opt in, to Anthropic with your own key.
It never reads your business records. Field definitions, yes; field contents, never.
And it won’t overrule you. It flags what departs from the guidance and asks “was that deliberate?” — you might have a great answer. Every finding passes a human review gate before it reaches a report. You keep or kill each one. The machine assesses; the architect decides.
Why grounding matters
The moment an assessment gets interesting is when a senior architect crosses their arms and says “says who?”
So every rule in orgspark declares its basis. Each threshold is labelled documented — with the Salesforce source cited — or opinionated, a practitioner default I admit is mine and you can tune. It renders as a “Method & Validation” sheet in the workbook. When someone challenges a verdict, the citation is right there.
That’s the difference between an assessment people argue with the evidence and one they wave away. I’d rather be told I’m wrong about a documented rule than be believed on a hunch.
The AI layer costs about a coffee
The engine needs zero AI. But bring your own Anthropic key and orgspark unlocks a judgment layer on top: a per-domain architecture-fitness deep pass, and reuse-vs-rebuild verdicts on every flagged component.
Every call is gated. Before anything sends, you see the exact payload, the token count and the worst-case cost. I ran the reuse pass across 50 components of my own org for 27 cents.
No key? There’s a “Judgment pack” you can paste into the claude.ai plan you already pay for. Your call, your spend, your review — every time.
It reads the tools you already love
orgspark isn’t a rival to sfdx-hardis or Org Check — it reads their output. Run them, drop the results in, and orgspark adds the judgment layer none of them claim. Nicolas Vuillamy’s sfdx-hardis and Christopher Ramm‘s OrgPulse both shaped how I thought about this; OrgPulse in particular proved the open, client-side, bring-your-own-AI model works.
What’s next
An org is thousands of decisions frozen in metadata — objects, fields, flows, agents, sharing, code — and most were never revisited. orgspark judges a growing slice today. The part it doesn’t yet judge isn’t hidden; it’s named, per org, in an “Unassessed Surface” list. Silence there means unassessed, never healthy.
That list is also the honest map of where this goes. The engine reads structure today — and structure is only the beginning of what’s frozen in an org.
Try it
You don’t need an org or an account to see what it does. There’s a bundled sample org seeded with 50+ real problems — a wide-open sharing rule, a dormant agent, a twelve-million-row object — so you can watch exactly what gets caught.
→ salesforcebinge.github.io/orgspark — click “Try with sample data.”
Want it on your real org? The site hands you a read-only script that describes your org, shows you every metadata type it found, and asks before it retrieves anything. Run it, drag the folder in, done.
Last launch, I shipped crmarcade — the tools layer, because tools are the most shareable part. This one is the assessment layer. Both are pieces of a bigger system I’ve been building on my architect journey, and there’s more to come.
Free. Open source. Built on personal time, for the community. If you’re an architect who’s ever squinted at an org and known — this one’s for you.
Until next time! 🙂
PS — Johann Furmann is driving something the ecosystem’s needed with CleanForce: org cleanliness as a real discipline. Different ends of the same mountain — Go follow his work.
