I opened two repos on GitHub: al-grano and brand-guide-kit, and why the rest stay private

Almost everything I build for the agency I build with Claude Code and Codex, and almost all of it stays private. Not because it is secret: most of those tools carry client data, paths from my computer or rules that only make sense in the way I work. Opened like that, they help nobody and expose somebody.

Every so often, though, a tool stops depending on all that. It works the same on someone else's computer, carries nothing from a client and I can explain it in one paragraph. That one I open. Today there are two, and in this post I tell you what each one does, who it is for, how to try it and what I check before opening the next one.

Why I open some and not others

On October 22 I counted the repos in the SVDIGINET organization on GitHub with gh repo list svdiginet: 11 repos, 2 public. The other 9 are private for one of three reasons: they hold client material, they depend on my personal setup, or I have not yet written the documentation someone who is not me would need.

The third reason is the most common and the most honest. A repo only I understand is not a public tool, it is my mess with the door open. Before I open one, its README has to answer three questions: what it does, what it does not do and how you try it without asking me for help.

How I counted: gh repo list svdiginet --json name,visibility,stargazerCount, run on Oct 22, 2026. The stars I quote below come from that same output. My personal account has its own public repos, but they are not agency tools and they are not part of this post.

al-grano: answers that get to the point

What it is. A response style for your coding assistant (Claude Code or Codex). It tells the assistant to answer first and explain after, cut the filler, number the steps and close a task in 10 lines or fewer. Code, commands and error messages stay exactly as they are.

Why I built it.I read AI answers all day. When each one opens with "great question" and ends with a summary of what I just read, I lose minutes hunting for the line that matters. I wrote it on September 18 for myself. This post was put together with al-grano on.

Who it is for. Anyone who uses a coding agent daily and wants less text to read without losing technical substance. It works in English and Spanish: it answers in the language you write in.

This is how an answer changes. It is the README example, written by us and translated here:

Great question. There are a few things going on with your build. The failing step is probably the type check, although it could also be the lint or a stale cache. You could clear the cache, run it again, and if that does not work check the tsconfig. Hope that helps you move forward!

Build fails at the type check: strict flag in tsconfig.json.
1. Fix the 3 errors it lists (or set "strict": false).
2. Run npm run build.

Aside: the cache is stale. Clear it later?

How it works inside. Two hooks. One runs when the session opens and loads the full rules; the other runs on every message you send and adds a short reminder, so the style survives when the conversation gets compacted. I ran them with an empty Claude config to measure how much text they add:

    1037

AL-GRANO MODE ACTIVE. Answer first (command/path/decision on line 1). Drop dead words. ...

(nothing: the flag was removed and the next message carries no reminder)

That is the real cost: about 1,037 words when the session opens and 33 words per message. In return, the answers come out shorter. I have not measured how much shorter in my sessions, so I will not give you a percentage. According to the repo, if the hooks fail they block nothing: the session opens normally, without the style.

How to try it.

1. Clone the repo into a folder of yours.
2. Follow INSTALL.md: in Claude Code two hooks go into ~/.claude/settings.json; in Codex you point your AGENTS.md at skills/al-grano/SKILL.md.
3. Open a new session and ask for something normal. To turn it off, type para al-grano, stop al-grano or normal mode.

github.com/svdiginet/al-grano →

In depth: al-grano's 11 rules, what it costs in context and how to install it in Claude Code and Codex.

brand-guide-kit: the brand, with evidence

What it is. A kit to write brand guides and audit the files behind them: the SVG logo, the colors and the live text. It brings a standard, templates (guide, audit and 3D spec), a skill for Claude Code and Codex, and two Python scripts with no third-party dependencies.

I built it because at the agency we hand brands to designers, to print shops and to whoever builds the website. The problem is never the PDF guide; it is that three months later someone uses a logo with a changed color or a font nobody approved, and nobody notices. I wanted a way to say "this file is the approved one" and prove it with one command. So it serves agencies, freelance designers and teams that hand a brand to someone else and need it to arrive intact. Also anyone using an AI agent to make brand pieces who wants the agent to check against written rules, not against what it remembers.

How it works. Each brand gets a brand.json file with its colors and its approved files, each with its SHA-256 fingerprint. The auditor compares: if someone changed one byte of the logo, or put a color in the CSS that is not in the palette, it flags it. I ran it on the sample brand the repo ships (fictional on purpose):

VERIFIED: mark.svg: File exists
VERIFIED: mark.svg: SHA256 matches declared baseline
VERIFIED: mark.svg: Explicit fill attributes match
VERIFIED: site.css: Hex literals use declared palette (not a computed-style audit)
UNKNOWN: visual: No visual fidelity, small-size, contrast, 3D or physical production certification
exit=2

Ran 14 tests in 0.150s

OK

My favorite line is the auditor's last one: UNKNOWN. The kit does not say your brand looks good. It says what it could check (the bytes, the declared colors) and what a script cannot check (how it looks small, the contrast, a printed piece). That is why it exits with code 2 on purpose: if you put it in an automation, you have to decide what to do with what could not be verified, instead of letting it pass silently.

What it does not do, and the README says so: it does not generate logos, it does not prove trademark clearance and it does not replace a physical print proof.

How to try it.

1. Clone the repo. All you need is Python 3.
2. Run python3 scripts/brand.py audit demos/sample-brand/brand.json and read the five lines.
3. Copy demos/sample-brand/brand.json next to your brand files, change the colors and files, and run the auditor again. Do not put a private brand.json inside a public repo.

github.com/svdiginet/brand-guide-kit →

In depth: brand-guide-kit's 9 gates, the brand.json contract and what each result means.

The two, side by side

If you are only going to try one, this table helps you pick:

What I check before opening a repo

For this post I ran both through the list I use before opening a repo again. I am sharing it because it works for any repo, yours or your agency's:

1. Nothing from a client. No names, no logos, no notes. That is why the brand-guide-kit sample brand is made up and says so in its own name.
2. Nothing from my computer. No keys, no tokens, no paths that only exist on my Mac. I check the full history, not just the latest version, because a deleted secret still lives in an old commit. On October 22 I ran gitleaks over the full history of both: zero findings.
3. A chosen license. Both are MIT. Without a license, a public repo can be read but not used with peace of mind.
4. A README that says what it does not do. It is what builds the most trust, and what gets forgotten the most.
5. A test that runs on its own. brand-guide-kit ships 14 tests and a check of its own documentation; al-grano is tested with the commands above.
6. A SECURITY.md file. It says how to tell us if someone finds a problem. Both have one.

If a repo fails the first or second point, it does not get opened, even if the code is fine. The rest can be fixed in an afternoon.

What I do not know yet

• How many people use them. One star each says nothing yet, and I have no other metric.
• How much shorter my answers get with al-grano. I feel it day to day, but I have not measured it.
• Which repo I open next. I have others in review against the list above; they come out when they pass all of it.

If you try one and something does not work, open an issue on GitHub. It is the fastest way to get it fixed, and the only way I know someone is using it.

If you want this in your business

These repos come out of the daily work at SVDIGINET, my local marketing agency in Central Florida: Google profiles, reviews, websites and automations for businesses here. If you would rather have us do it than clone a repo, the details are at svdiginet.com. And if you want the story of why the agency exists, it is in this post.

Sources: github.com/svdiginet, checked Oct 22, 2026; al-grano; brand-guide-kit