Data

Tools

Indexes

Applications

Experiments

Awesome

An open API service indexing awesome lists of open source software.

https://github.com/whystrohm/sample-foundrkit

Sample Foundrkit for Linear. The public reference implementation of the format. Real rules, real linter, real CI.
https://github.com/whystrohm/sample-foundrkit

Last synced: 14 days ago
JSON representation

Sample Foundrkit for Linear. The public reference implementation of the format. Real rules, real linter, real CI.

Awesome Lists containing this project

README 

          
# Sample Foundrkit for Linear



A working public Foundrkit for [linear.app](https://linear.app/), built as a reference implementation for the format. Real rules, real linter, real CI. The Action on this repo blocks any PR that violates the Foundrkit.



The bad-PR demonstration lives at [`PR #1`](https://github.com/whystrohm/sample-foundrkit/pull/1), an intentionally-slop draft that fails CI. Click through to see the workflow log catch the violations in plain text.



> If you want a Foundrkit like this generated for your own brand, the engine that produces it lives at [whystrohm.com](https://whystrohm.com).



---



## What is a Foundrkit?



A Foundrkit is the language constants of a brand, captured as code instead of a PDF. Four files plus an example library, all rendered into formats a machine reads on every output.



The problem the format solves: every founder-led brand has a style guide. Almost none of them get used. The guide is a PDF in Notion. The contractor opens ChatGPT and writes whatever ChatGPT writes. The output ships. The brand sounds like every other AI-written SaaS company.



The Foundrkit flips that. The same rules live as `brand.config.ts` (the voice constants), `foundrkit.rules.json` (the structural rules), `forbidden.json` (the blocklist), and `CLAUDE.md` (the operating instructions for any AI writing assistant pointed at the repo). A CI job runs the rules against every PR. Nothing on-brand slips through and nothing off-brand merges.



A real engagement to build one of these for a brand takes about 30 days. This repo is the artifact a buyer gets at the end. Same shape, populated with their language instead of Linear's.



## What's in this repo



```

sample-foundrkit/

├─ README.md                    ← you're here

├─ brand.config.ts              ← voice constants: paragraph, 7 axes,

│                                  register, approved patterns

├─ foundrkit.rules.json         ← 6 structural rules (block + warn)

├─ forbidden.json               ← blocklist: 50+ words and phrases,

│                                  categorized

├─ CLAUDE.md                    ← operating instructions for Claude Code

│                                  / any AI writing assistant

├─ foundrkit-lint.mjs           ← the linter the CI Action runs.

│                                  Zero dependencies, runs on Node 20+

├─ CODEOWNERS                   ← every PR routes to @whystrohm

├─ examples/

│  ├─ good-post.md              ← Linear-flavored draft that PASSES

│  └─ bad-post.md               ← deliberately slop draft that FAILS

└─ .github/workflows/

   └─ foundrkit-lint.yml        ← runs the linter on every PR

```



Everything is human-readable. Click any file in the GitHub UI.



## How does the firewall use it?



The Action in [`.github/workflows/foundrkit-lint.yml`](.github/workflows/foundrkit-lint.yml) runs on every pull request. It checks out the branch, finds changed markdown files, and runs:



```

node foundrkit-lint.mjs 

```



The linter reads `foundrkit.rules.json` and `forbidden.json` and applies them to each file. Output looks like this in the CI log:



```

examples/bad-post.md

  ✗ no-em-dashes               em-dash on line 3 col 47

  ✗ no-forbidden-words         "game-changing" (globally_banned_words) on line 1 col 14

  ✗ no-forbidden-words         "thrilled" (enthusiasm_verbs) on line 3 col 4

  ...

  ⚠ max-sentence-length        38% of sentences over 25 words (limit 30%)



1 file checked, 7 block violations, 1 warning. FAILED.

```



Block rules (no em-dashes, no forbidden words) fail CI with exit code 1. Warn rules surface in the log but the build passes. The two block categories are non-negotiable. They're the patterns that mark text as AI slop.



## See it work



There's a permanent open pull request on this repo that demonstrates the firewall catching a slop draft:



**[`PR #1`: Firewall demo, slop draft fails the Foundrkit](https://github.com/whystrohm/sample-foundrkit/pull/1)**



The PR adds `examples/bad-post.md` with deliberate violations. The Action runs and fails. The PR is not mergeable. It stays open forever as the demonstration.



![foundrkit-lint Action failed on PR #1, 40 violations, status Failure](./firewall-demo-failed-action.png)



What the CI log shows (excerpted):



```

examples/bad-post.md

  ✗ no-em-dashes                 em-dash on line 1 col 27

  ✗ no-em-dashes                 em-dash on line 3 col 69

  ✗ no-em-dashes                 em-dash on line 5 col 131

  ✗ no-em-dashes                 em-dash on line 20 col 171

  ✗ no-forbidden-words           "game-changing" (globally_banned_words) on line 1 col 31

  ✗ no-forbidden-words           "transformative" (globally_banned_words) on line 3 col 73

  ✗ no-forbidden-words           "cutting-edge" (globally_banned_words) on line 3 col 89

  ✗ no-forbidden-words           "Artificial intelligence is fundamentally reshaping" (ai_era_clutter) on line 5 col 1

  ✗ no-forbidden-words           "purpose-built system" (abstract_noun_clusters) on line 5 col 135

  ✗ no-forbidden-words           "shared, structured environment" (abstract_noun_clusters) on line 5 col 201

  ✗ no-forbidden-words           "revolutionary" (globally_banned_words) on line 7 col 28

  ✗ no-forbidden-words           "innovative" (globally_banned_words) on line 9 col 20

  ✗ no-forbidden-words           "robust" (globally_banned_words) on line 9 col 32

  ✗ no-forbidden-words           "seamless" (globally_banned_words) on line 9 col 44

  ✗ no-forbidden-words           "harness the power of" (ai_era_clutter) on line 9 col 89

  ✗ no-forbidden-words           "holistic" (globally_banned_words) on line 9 col 158

  ✗ no-forbidden-words           "ecosystem" (globally_banned_words) on line 9 col 243

  ✗ no-forbidden-words           "leverage" (globally_banned_words) on line 13 col 3

  ✗ no-forbidden-words           "relentless focus" (abstract_noun_clusters) on line 15 col 34

  ✗ no-forbidden-words           "fast execution" (abstract_noun_clusters) on line 15 col 55

  ...

  ⚠ max-sentence-length          33% of sentences over 25 words (limit 30%)



1 file checked, 40 block violations, 1 warning. FAILED.

```



40 block violations across 5 forbidden-word categories. The build exits 1. The PR is unmergeable until the violations are fixed.



The merged `main` branch has [`examples/good-post.md`](./examples/good-post.md), a Linear-flavored post on the same topic (Linear Agent GA) that passes the same firewall.



Open both posts side-by-side. The difference between the brand sounding like itself and the brand sounding like every other SaaS company is in the structural choices the Foundrkit enforces, not in the topic.



## Run the linter locally



```bash

git clone https://github.com/whystrohm/sample-foundrkit.git

cd sample-foundrkit

node foundrkit-lint.mjs examples/good-post.md   # exit 0

node foundrkit-lint.mjs examples/bad-post.md    # exit 1

```



Node 20+ required. No `npm install` step. The linter has zero runtime dependencies.



## Want one for your brand?



The engine that generated this Foundrkit for Linear runs at [whystrohm.com](https://whystrohm.com). Paste your own URL on the homepage and you'll see what a Foundrkit preview looks like for your brand. The build engagement that produces a real Foundrkit repo (this shape, populated with your language) is the next step.