Herd Worktree

Laravel Herd Worktree

Spin up an isolated, runnable copy of a Laravel project on its own Herd site for feature-branch work - without colliding with your main checkout or, on Herd Pro, your shared dev database.

Announce at start: "Using laravel-herd-worktree to set up an isolated Laravel + Herd workspace."

Design: this is a thin Herd layer on top of vd:worktree

Do not reinvent worktree mechanics. vd:worktree already creates the worktree under .worktrees/<repo>-<feature>/, copies all .env* (nested included), assigns a deterministic 10-port block (.env.worktree → WORKTREE_PORT_BASE), keeps .git/info/exclude clean, anchors agent artifacts to the main checkout, and enters the worktree. This skill only adds what Herd needs and then defers finishing to vd:ship / vd:git.

Prerequisites

• Laravel Herd (Pro adds shared MySQL/Postgres/Redis services - relevant to DB isolation below). macOS.
• Vite + npm frontend (adjust if yarn/pnpm).
• Laravel Sanctum only if the project uses it (steps below are conditional).

Setup

1. Create the worktree via vd:worktree

Detect the branch (ticket key is authoritative), then:

WT="$(for d in "$HOME/skills/skills/worktree" "$HOME/.claude/skills/worktree" "$HOME/.agents/skills/worktree"; do [ -d "$d" ] && { echo "$d/scripts/worktree.cjs"; break; }; done)"
node "$WT" create "<branch>" --no-prefix --json

$WT resolves to the first worktree install that exists - the $HOME/skills dev repo, else the Claude (~/.claude/skills) or Codex (~/.agents/skills) install; pick it once per session and reuse it below. Parse the JSON for the worktree path and portBase, then move the session into the worktree (per the sessionSwitch block). The .env is already copied - you only rewrite it below.

PROJECT_NAME=$(basename "$(git rev-parse --show-toplevel)")   # main repo name
SANITIZED_BRANCH=$(echo "$BRANCH_NAME" | tr '/' '-')
SITE_NAME="$PROJECT_NAME-$SANITIZED_BRANCH"                    # matches .worktrees/<repo>-<feature>

2. Link to Herd - match the source site's scheme

Detect whether the source project is secured (don't assume HTTP):

herd links | grep -E "\b$PROJECT_NAME\b"     # SSL column, or:
grep '^APP_URL=' /path/to/main/.env          # https:// ⇒ secured

cd <worktree-path>
herd link "$SITE_NAME"
# If the source is HTTPS (Herd Pro default for team projects):
herd secure "$SITE_NAME"     # ⇒ https://$SITE_NAME.test
# Only if the source is plain HTTP: skip herd secure (HTTP matches the Vite dev server)

Forcing HTTP on a project that is normally HTTPS causes scheme/cookie mismatches. Mirror the source.

3. Rewrite the worktree .env (conditionally)

SCHEME = https if secured, else http. Edit only keys that exist in the source .env:

ENV=<worktree-path>/.env
sed -i '' "s|^APP_URL=.*|APP_URL=$SCHEME://$SITE_NAME.test|" "$ENV"

# SESSION_DOMAIN: only set if the source set a real domain (skip when it's null/empty - single-host).
grep -qE '^SESSION_DOMAIN=.+' /path/to/main/.env && [ "$(grep '^SESSION_DOMAIN=' /path/to/main/.env)" != "SESSION_DOMAIN=null" ] \
  && sed -i '' "s|^SESSION_DOMAIN=.*|SESSION_DOMAIN=$SITE_NAME.test|" "$ENV"

# Sanctum: only if present.
grep -q '^SANCTUM_STATEFUL_DOMAINS=' "$ENV" \
  && sed -i '' "s|^SANCTUM_STATEFUL_DOMAINS=\(.*\)|SANCTUM_STATEFUL_DOMAINS=\1,$SITE_NAME.test|" "$ENV"

# Secure-cookie flag only matters for HTTP.
[ "$SCHEME" = "http" ] && echo "SESSION_SECURE_COOKIE=false" >> "$ENV"

4. Database - isolate or share (Herd Pro footgun)

On Herd Pro the worktree's copied .env points at the same shared DB as main. A feature branch running migrate / migrate:fresh / seeders would mutate your main dev database. Ask (AskUserQuestion in Claude Code; plain-text question elsewhere):

AskUserQuestion:
  question: "This worktree will run migrations against DB_DATABASE. Isolate it?"
  header: "Database"
  options:
    - label: "Isolate (Recommended)"  → create a per-worktree DB, rewrite DB_DATABASE
    - label: "Share main DB"          → reuse the same DB (read-only / no migrations)

If Isolate (MySQL example; mirror for Postgres):

DBNAME=$(grep '^DB_DATABASE=' "$ENV" | cut -d= -f2)_$SANITIZED_BRANCH
mysql -h 127.0.0.1 -u "$(grep '^DB_USERNAME=' "$ENV" | cut -d= -f2)" -e "CREATE DATABASE IF NOT EXISTS \`$DBNAME\`"
sed -i '' "s|^DB_DATABASE=.*|DB_DATABASE=$DBNAME|" "$ENV"
# (migrate in step 6)

5. Vite - scheme-aware

ls <worktree-path>/vite.config.ts <worktree-path>/vite.config.js 2>/dev/null   # detect either

• HTTPS site: laravel-vite-plugin auto-detects the Herd TLS cert created by herd secure and serves Vite over HTTPS on the right host - leave it to the plugin. Add server: { cors: true } only if you actually see cross-origin blocks.
• HTTP site: ensure server: { host: 'localhost', cors: true } (do not use 0.0.0.0).

6. Register teardown hook (so vd:worktree clean/remove cleans Herd + DB)

vd:worktree runs .worktree/hooks/pre-remove inside the worktree on both remove and clean (env exported: WORKTREE_NAME, WORKTREE_PATH, WORKTREE_SOURCE, …). Write one at setup so the Herd site and any isolated DB are never orphaned - worktree clean then handles Laravel teardown for free:

mkdir -p <worktree-path>/.worktree/hooks
cat > <worktree-path>/.worktree/hooks/pre-remove <<'EOF'
#!/usr/bin/env bash
# Herd + DB teardown - runs on `vd:worktree remove` and `clean`. Failure warns, never blocks.
set -u
herd unlink "$WORKTREE_NAME" 2>/dev/null || true
pkill -f "node.*vite.*$WORKTREE_NAME" 2>/dev/null || true
# Drop the DB only if this worktree was isolated (its DB differs from the source's).
DB=$(grep -m1 '^DB_DATABASE=' "$WORKTREE_PATH/.env" 2>/dev/null | cut -d= -f2)
SRC_DB=$(grep -m1 '^DB_DATABASE=' "$WORKTREE_SOURCE/.env" 2>/dev/null | cut -d= -f2)
if [ -n "${DB:-}" ] && [ "$DB" != "${SRC_DB:-}" ]; then
  mysql -h 127.0.0.1 -u root -e "DROP DATABASE IF EXISTS \`$DB\`" 2>/dev/null || true
fi
EOF
chmod +x <worktree-path>/.worktree/hooks/pre-remove

The DB drop is guarded - it never touches the shared/source database, only a per-worktree one created in step 4.

7. Install, migrate, run

cd <worktree-path>
composer install --no-interaction          # worktrees don't share vendor/
npm install                                # …or node_modules/
php artisan config:clear && php artisan cache:clear
[ -n "$DBNAME" ] && php artisan migrate     # only when DB was isolated; add --seed if wanted

pkill -f "node.*vite" 2>/dev/null; rm -f public/hot   # free a stale Vite/hot file
npm run build     # or keep `npm run dev` running while browsing through Herd

Site: $SCHEME://$SITE_NAME.test.

Finishing - delegate, don't reinvent

When work is done, route to the existing skills rather than a bespoke PR flow:

• Ship it: vd:ship (test → review → version → PR) or vd:git for a single conventional commit + push from the worktree branch.
• Bring changes into main checkout: git merge <branch> --no-commit --no-ff from the main tree, review, commit.

Then clean up through vd:worktree - the pre-remove hook from step 6 does the Herd unlink + isolated-DB drop automatically:

node "$WT" remove "$SITE_NAME"   # one worktree: hook → branch → metadata
# or sweep all merged/stale worktrees at once (each runs its own teardown hook):
node "$WT" clean --yes

Because teardown lives in the worktree's pre-remove hook, vd:worktree clean tears down the Herd site and isolated DB for every swept worktree - no orphaned .test sites or leftover databases. (Stop Vite first if it's still running: pkill -f "node.*vite".)

Herd Pro notes

• Shared services: Pro's MySQL/Postgres/Redis are shared across all sites - DB isolation (step 4) is the main safeguard; Redis/queue can collide too (use a per-worktree REDIS_PREFIX / DB_REDIS if the branch hits queues).
• HTTPS is the Pro default for team projects - mirror it (steps 2-3), don't force HTTP.
• Per-site PHP: herd isolate <php-version> inside the worktree if it needs a different PHP than the global default.

Common issues

Checklist

• Worktree created via vd:worktree (.worktrees/$SITE_NAME, .env copied, ports assigned)
• herd link done; scheme matches source (secured ⇒ herd secure)
• .env rewritten: APP_URL (correct scheme), SESSION_DOMAIN/Sanctum only if applicable
• Database decision made (isolated DB + DB_DATABASE rewritten, or explicitly sharing)
• .worktree/hooks/pre-remove written (Herd unlink + isolated-DB drop on worktree remove/clean)
• vite.config.{ts,js} correct for the scheme
• composer install + npm install done; npm run build done or npm run dev running; migrations run if isolated
• Site reachable at $SCHEME://$SITE_NAME.test

CRITICAL: working directory after setup

All subsequent work (edits, artisan, tests, git) MUST use the worktree path. vd:worktree enters it by default; stay there. The main checkout is a different copy - editing it changes the wrong files.

Integration

• Composes: vd:worktree (mechanics), vd:ship / vd:git (finishing).
• Pairs with: vd:cook / vd:fix (implement in the worktree).