Atlas / Skills / jezweb / Fork Discipline

Fork DisciplineSAFE

skills/jezweb/fork-discipline

Skills for Claude Code CLI such as full stack dev Cloudflare, React, Tailwind v4, and AI integrations.

Verdict
SAFE
Grade
B
Trust score
89 /100
Version
—
Hosts
—
License
MIT
Stars
1,059
01

Overview

Skills for Claude Code CLI such as full stack dev Cloudflare, React, Tailwind v4, and AI integrations.

Read from source at commit be410003e718OBSERVED · 2026-10-12
02

What it tells the agent

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: fork-discipline
description: "Audit and enforce the core/client boundary in multi-client projects. Detects where shared platform code is tangled with client-specific code, finds hardcoded client checks, config files that replace instead of merge, scattered client code, migration conflicts, and missing extension points. Produces a boundary map, violation report, and refactoring plan. Optionally generates FORK.md documentation and restructuring scripts. Triggers: 'fork discipline', 'check the boundary', 'is this core or client', 'platform audit', 'client separation', 'fork test', 'refactor for multi-client', 'clean up the fork'."
compatibility: claude-code-only
allowed-tools:
  - Read
  - Write
  - Edit
  - Glob
  - Grep
  - Bash
---

# Fork Discipline

Audit the core/client boundary in multi-client codebases. Every multi-client project should have a clean separation between shared platform code (core) and per-deployment code (client). This skill finds where that boundary is blurred and shows you how to fix it.

## The Principle

```
project/
  src/            ← CORE: shared platform code. Never modified per client.
  config/         ← DEFAULTS: base config, feature flags, sensible defaults.
  clients/
    client-name/  ← CLIENT: everything that varies per deployment.
      config      ← overrides merged over defaults
      content     ← seed data, KB articles, templates
      schema      ← domain tables, migrations (numbered 0100+)
      custom/     ← bespoke features (routes, pages, tools)
```

**The fork test**: Before modifying any file, ask "is this core or client?" If you can't tell, the boundary isn't clean enough.

## When to Use

- Before adding a second or third client to an existing project
- After a project has grown organically and the boundaries are fuzzy
- When you notice `if (client === 'acme')` checks creeping into shared code
- Before a major refactor to understand what's actually shared vs specific
- When onboarding a new developer who needs to understand the architecture
- Periodic health check on multi-client projects

## Modes

| Mode | Trigger | What it produces |
|------|---------|-----------------|
| **audit** | "fork discipline", "check the boundary" | Boundary map + violation report |
| **document** | "write FORK.md", "document the boundary" | FORK.md file for the project |
| **refactor** | "clean up the fork", "enforce the boundary" | Refactoring plan + migration scripts |

Default: **audit**

---

## Audit Mode

### Step 1: Detect Project Type

Determine if this is a multi-client project and what pattern it uses:

| Signal | Pattern |
|--------|---------|
| `clients/` or `tenants/` directory | Explicit multi-client |
| Multiple config files with client names | Config-driven multi-client |
| `packages/` with shared + per-client packages | Monorepo multi-client |
| Environment variables like `CLIENT_NAME` or `TENANT_ID` | Runtime multi-client |
| Only one deployment, no client dirs | Single-client (may be heading multi-client) |

If single-client: check if the project CLAUDE.md or codebase suggests it will become multi-client. If so, audit for readiness. If genuinely single-client forever, this skill isn't needed.

### Step 2: Map the Boundary

Build a boundary map by scanning the codebase:

```
CORE (shared by all clients):
  src/server/          → API routes, middleware, auth
  src/client/          → React components, hooks, pages
  src/db/schema.ts     → Shared database schema
  migrations/0001-0050 → Core migrations

CLIENT (per-deployment):
  clients/acme/config.ts    → Client overrides
  clients/acme/kb/          → Knowledge base articles
  clients/acme/seed.sql     → Seed data
  migrations/0100+          → Client schema extensions

BLURRED (needs attention):
  src/server/routes/acme-custom.ts  → Client code in core!
  src/config/defaults.ts line 47    → Hardcoded client domain
```

### Step 3: Find Violations

Scan for these specific anti-patterns:

#### Client Names in Core Code

```bash
# Search for hardcoded client identifiers in shared code
grep -rn "acme\|smith\|client_name_here" src/ --include="*.ts" --include="*.tsx"

# Search for client-specific conditionals
grep -rn "if.*client.*===\|switch.*client\|case.*['\"]acme" src/ --include="*.ts" --include="*.tsx"

# Search for environment-based client checks in shared code
grep -rn "CLIENT_NAME\|TENANT_ID\|process.env.*CLIENT" src/ --include="*.ts" --include="*.tsx"
```

**Severity**: High. Every hardcoded client check in core code means the next client requires modifying shared code.

#### Config Replacement Instead of Merge

Check if client configs replace entire files or merge over defaults:

```typescript
// BAD — client config is a complete replacement
// clients/acme/config.ts
export default {
  theme: { primary: '#1E40AF' },
  features: { emailOutbox: true },
  // Missing all other defaults — they're lost
}

// GOOD — client config is a delta merged over defaults
// clients/acme/config.ts
export default {
  theme: { primary: '#1E40AF' },  // Only overrides what's different
}
// config/defaults.ts has everything else
```

Look for: client config files that are suspiciously large (close to the size of the defaults file), or client configs that define fields the defaults already handle.

**Severity**: Medium. Stale client configs miss new defaults and features.

#### Scattered Client Code

Check if client-specific code lives outside the client directory:

```bash
# Files with client names in their path but inside src/
find src/ -name "*acme*" -o -name "*smith*" -o -name "*client-name*"

# Routes or pages that serve a single client
grep -rn "// only for\|// acme only\|// client-specific" src/ --include="*.ts" --include="*.tsx"
```

**Severity**: High. Client code in `src/` means core is not truly shared.

#### Missing Extension Points

Check if core has mechanisms for client customisation without modification:

| Extension point | How to check | What it enables |
|----------------|-----------
03

Trust audit

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

LayerWhat it checksResult
L0Provenance & inventoryPASS
L1Static analysis of the codeNA
L2Instruction surface (what it tells the agent)PASS
L3Class-specific surfacePASS
L4Behavioural (sandbox)SKIPPED

What the source does

Filesystem
none-observed
Network
none-observed
Shell
none-observed
Dependencies
pinned
Secrets in source
none-found

Findings (0)

No findings outside the package's declared scope.

Gates applied: no_behavioural_pass.

Audited 2026-10-12 · audit v0.4.1 · source sha be410003e718full audit observations/trust-audit/skill/jezweb__fork-discipline.json · Report an issue / request a re-scan
04

Audit history

Every audit this skill has had.

DateSourceVerdictGradeScoreChange
2026-10-12be410003e718SAFEB89first audit
05

Questions

What does the Fork Discipline skill do?

Skills for Claude Code CLI such as full stack dev Cloudflare, React, Tailwind v4, and AI integrations.

Is Fork Discipline safe to install?

The audit found nothing in the source that contradicts what it says it does, and graded it B (89/100). Grade A is held back for packages that have also passed a sandboxed behavioural run, which is why a clean skill reads B.

What can Fork Discipline access on my machine?

The audit observed no filesystem, network or shell use at all in its source.

How current is this page?

The grade is for one exact copy of the source (be410003e718), read on 2026-10-12. The repository is watched, and a new audit runs when it changes — this is the first audit.

Advertisement