Skip to main content

Featured

Plumbing Business Startup Cost: I Rebuilt the 1-to-5 Van Ladder From Its Own 19 Rows

The short version: "How much does it cost to start a plumbing business" has no single answer, and a kit that gives you one number is lying by format. So I unzipped this one, rebuilt its one-to-five-van ladder from its own 19 line items in Python, re-priced all five worked calls from their inputs, and ran its checker. The ladder footed to the cent, the five calls came back identical, and the one honest thing the headline number hides is that half of it is cash you keep, not money you spend. Every row of the ladder footed when I rebuilt it from the 19 line items, and the break-even landed on 52.53 calls — which is 53 whole calls, not 52. The problem: a trade business with no written prices If you are a service plumber or a lead tech about to go owner-operator, the skills are not the problem. You can read a leak, trace a drain, sweat a joint and leave a clean cabinet behind. Almost nobody in that position has ever priced a job on paper. So the first quotes are guesses. A nu...

RulePort Review: I Ran Its Installer Six Ways. Four Claims Held, Two Leaked.

The short version: a rule pack that claims it "compiles one canonical rule set into Cursor, Claude Code, Windsurf and Codex" is only worth the tooling behind it. So I unzipped RulePort into an empty directory and ran its installer six ways — dry-run, real, twice for idempotency, against foreign files, with no agents present, and with deliberate drift. Four of those behaved exactly as advertised. Two leaked. Both leaks are below, with the output.

The problem this thing is trying to solve

You are probably running more than one coding agent. The 2026 setup for a working developer is Cursor in one repo, Claude Code in the terminal, Windsurf for the client codebase that standardised on it, and Codex or Copilot CLI for a side project. Switching is fine. What is not fine is that the rules you spent months learning to write do not travel.

Those rules are the boring discipline stuff: run the real checks before declaring done, write the test first, never swallow an exception, review your own diff before shipping it. None of it is clever. All of it is the reason agent-written code is usable at all. And every tool wants it in a different place:

ToolWhere it reads rulesFormat
Cursor.cursor/rules/*.mdcYAML frontmatter + globs
Claude Codeskills/*/SKILL.md, CLAUDE.md, slash commandsfrontmatter with a "Use when…" trigger
Windsurf.windsurf/rules/*.mddescription + pattern frontmatter
Codex, OpenCode, Copilot CLI, Aiderone AGENTS.md at the project rootplain Markdown

Four incompatible destinations for one set of thinking. Your .cursorrules does not become a Claude Code skill; you re-author it. So the honest options today are all bad: maintain it by hand in three places and let the versions drift, paste from a free awesome-list that went months without an update, or buy a file dump from Gumroad that ships without a compiler, an installer, or any way to know whether what is on disk still matches what you bought.

The specific thing that goes wrong with a file dump is quiet. Your agent keeps following a rule you corrected three updates ago, and nothing in the toolchain tells you. That is a drift problem, and drift problems need tooling, not more files.

What is actually in the pack

RulePort is 12 canonical rule files plus a small Python compiler that emits them into each runtime's native format, a bash installer that detects which agents you actually have, and a diff command. I unzipped it and counted, rather than reading the sales page:

  • 58 files, 174,768 bytes uncompressed (103,003 bytes as the zip you download).
  • 12 canonical rules in canonical/ — the single source of truth.
  • 39 compiled files: 12 Cursor .mdc, 14 Claude Code files (12 skills + CLAUDE.md + one /ruleport-status command), 12 Windsurf .md, 1 AGENTS.md (24,336 bytes carrying all 12 rules).
  • tools/ruleport.py — 457 lines, and its only imports are argparse, difflib, pathlib, shutil, sys. Standard library only, so no pip install and nothing to audit.
  • install.sh (111 lines), docs/customization.md, docs/update-policy.md, LICENSE, CHANGELOG.md.

The 12 rules are verify-before-done, test-first, error handling, types and docstrings, security baseline, self code review, database safety, API design, dependency hygiene, git hygiene, frontend UI and accessibility, and documentation updates. They are not clever rules. They are the ones that stop the classic agent failures — the confident "done" with untested code, the swallowed exception, the dependency added for a problem the standard library already solved.

What I measured, command by command

Every line below came from a clean copy unzipped into an empty directory, run with an empty working project and, where noted, a stripped HOME. The full transcripts are in my notes; this is the summary.

1. The compiler is genuinely reproducible

python3 tools/ruleport.py compile printed compiled 12 rules -> 39 files. I hashed the shipped compiled/ tree, ran compile, hashed again, and ran it a third time: 0 files changed against the shipped output, 0 changes between run one and run two. That matters more than it sounds — a "compiler" that reorders or reformats its own output on every run makes every commit look dirty and defeats the drift check. This one is byte-stable.

2. --dry-run writes nothing, and that is verifiable

bash install.sh --dry-run listed every file it would create. I hashed the target project directory before and after: zero files. The dry-run is real, not a verbose mode that writes and tells you afterwards.

3. The installer is idempotent

Real install into an empty project with install --agent all --target wrote 39 files, the exact compiled set. Running the identical command again reported every file as unchanged, and the sha256 map of the tree was identical. Nothing piled up, nothing was duplicated, nothing gained a second marker line.

4. It refuses to clobber your own files — without --force

This is the claim I most expected to be decorative. I planted a hand-written CLAUDE.md and AGENTS.md (no RulePort marker) in the project, then ran the full install. Both were skipped and left byte-identical (# my own file on line one, still there). Re-running with --force replaced them with the generated versions, and diff then reported in sync. That is the correct default: an installer that eats a CLAUDE.md you wrote yourself is worse than no installer.

5. The universal fallback fires

With a fresh empty HOME and a PATH that exposes no agent binaries at all, install.sh printed no runtimes detected — installing the AGENTS.md fallback and wrote exactly one file: ./AGENTS.md. The claim in the README is true on a machine with nothing installed.

6. diff detects drift — and its exit code means what it says

Fresh install: diff printed in sync — installed files match the compiled pack, exit 0. I appended one line to a single Cursor rule file and re-ran: exit 1, with a real unified diff naming the file and the extra line. Bad arguments also fail loudly rather than silently: an unknown --agent exits 2 with the valid choices, an unknown subcommand exits 2, an unknown shell flag exits 2 with usage. I would rather a tool shout at me than guess.

Where it leaks — and you should know this before paying

Two things did not survive contact with a real machine.

Leak one: diff can report "in sync" over zero files. Run diff in a project where nothing is installed — or against a directory where the runtimes were never installed — and it prints one line per runtime saying "not installed at … — skipped", then "in sync — installed files match the compiled pack", exit 0. All four runtimes skipped, zero files compared, green tick. If you wire the exit code into CI as the README suggests, a machine where the install silently never happened will pass the drift gate. Read the comparing N file(s) lines, not just the exit code, or assert on a file count you expect.

Leak two: "claude detected" means "a claude binary is on your PATH". The detection is [ -d "$HOME/.claude" ] || command -v claude. On this machine the Claude CLI is installed but Claude Code is not something I use in every project — the installer still wrote 14 files into ~/.claude, globally, from a project directory, without asking. Note the scoping surprise too: the README's table says Claude Code installs to ~/.claude/ (global) or .claude/ (per project), but bash install.sh only ever does the global one. To get a project-local .claude/ you must call python3 tools/ruleport.py install --agent claude --target <dir> instead. Confirmed by doing both: the shell script targeted $HOME/.claude, the --target form wrote into the project.

Neither leak is fatal and neither is dishonest — the behaviour is documented loosely enough to be defensible. But "green drift check" that checked nothing is exactly the class of false confidence this pack exists to prevent, so it belongs in the review, not in a footnote.

Who it is NOT for

I would rather you skip it than regret it:

  • Not for you if you run exactly one agent. A compiler that emits four formats is overhead when you only consume one. Hand-write your rules and save the dollar.
  • Not for you if your team already has a maintained docs or conventions tree that agents read. You have a source of truth; you do not need a second one.
  • Not for anyone who needs a drift gate they can trust blind. Because of leak one above, you must assert on the compared-file count, not the exit code alone.
  • Not for teams who want rules generated from source code. This ships opinionated prose, not a linter and not an analyser.
  • Not for you if you need Windows-native tooling. The compiler is pure Python and portable, but the installer is bash.
  • Not for anyone expecting Cursor, Claude Code, Windsurf or Codex to behave identically once the rules land. Each runtime loads its format on its own terms — a .mdc glob is not a Skill trigger. RulePort standardises the text, not the loaders.

What I would do differently

Nothing in the pack. I do not own this listing's contents and I did not change them. What I changed was how I run it: install with an explicit --target instead of a bare bash install.sh when I care about project scope, and assert on "comparing 12 file(s)" rather than trusting a green exit code. Both of those are one-line habits, and both are the kind of thing the tool is about.

Honesty block

Example figures are illustrative planning inputs — not market rates and not earnings. Nothing here is legal, tax, insurance, licensing, veterinary, automotive-repair, or financial advice. No claims are made about CRM, map, API, or invoicing integrations that this bundle does not ship — it ships 12 canonical rule files, a 457-line Python compiler, a 111-line bash installer and two docs, as counted above. No ratings, reviews, testimonials, or sales figures are quoted in this article, because I have none that would be honest to quote: the live page shows a rating count of zero.

Price and the link

Disclosure: this post promotes a product we made. The sale is ours directly — this is not an affiliate link and there is no third-party commission on it. I am writing it because the measured behaviour above is worth having in public before anyone spends money, not because a commission depends on it.

The pack is on Gumroad at https://nandminc.gumroad.com/l/xcbqxr for $1 intro price (the page also displays SAR 3.75 as a local-currency estimate). For that you get the 58-file zip, the compiler, the installer, the two docs and free updates when the agent formats change, delivered through your Gumroad receipt link. Questions first? Email nandminc2020@gmail.com.

License in one line: personal and commercial use is allowed, including inside organisations and for client work; resale or redistribution of the pack or its files as a standalone product, or inside another rules/skills collection, is not.

FAQ

Does this replace my .cursorrules? Effectively yes. Modern Cursor reads .cursor/rules/*.mdc, which supports per-file glob scoping the legacy single file cannot. Keep the old file until you are happy, then delete it.

Do I need to install anything? Python 3.8+ and bash. No pip packages, no network access, nothing running in the background. The compiler imports only the standard library, as counted above.

Can my whole team use one purchase? Yes — personal and commercial use is permitted, including client work. Redistributing the pack itself is not.

What happens when Cursor or Windsurf change their format? You recompile against an updated compiler. That is the whole design: a format change should be a recompile, not a rewrite of your rules. Details are in docs/update-policy.md inside the pack.

If you have a multi-agent setup and a rule-drift story of your own, I want to read it: nandminc2020@gmail.com. For more AI tooling deep-dives on AI Stack Lab, start at aistacklab2026.blogspot.com.

Comments