Menu
Load project context and show current status. Use at the start of a session or when context is needed.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Basé sur la classification professionnelle SOC
Cross-project audit and sync. Backs up, bootstraps, promotes, syncs, and verifies all downstream projects.
Structured pre-planning research. Explores codebase, asks clarifying questions, and produces a brainstorm output file for /plan-feature input.
Capture knowledge from development sessions. Debug patterns, architecture decisions, framework gotchas, and integration learnings compound over time.
Run parallel code reviews using specialized agents (security, performance, simplicity, nextjs-react). Produces a structured report.
Initialize a new project from Agent Kit boilerplate. Use when creating a new downstream project.
Pre-Production Security & Quality Check with Strix AI. Use BEFORE deploying to production to catch vulnerabilities and issues.
| name | prime |
| description | Load project context and show current status. Use at the start of a session or when context is needed. |
| disable-model-invocation | true |
| allowed-tools | Read, Bash, Glob, Write, AskUserQuestion |
Build comprehensive understanding of the codebase AND show current project status so work can resume immediately.
You are operating inside the Lucid Labs Agent Kit.
When rendering the session start screen, command list, or skill overview:
Standard Claude Code commands remain available (e.g. /help, /clear, /config, etc.) They are just not listed in the Agent Kit session commands block.
Deviation from the defined Agent Kit command set is considered an error.
KRITISCH: Bevor irgendetwas anderes passiert, MUSS der Kontext erkannt werden.
# 1. Aktuelles Verzeichnis prüfen
pwd
# 2. Prüfen ob PROJECT-CONTEXT.md existiert
cat .claude/PROJECT-CONTEXT.md 2>/dev/null
Kontext-Entscheidungsbaum:
Existiert .claude/PROJECT-CONTEXT.md?
├── JA → Lese type: Feld
│ ├── type: downstream → DOWNSTREAM PROJEKT
│ │ └── Lese active_project.name und active_project.prd
│ └── type: upstream → UPSTREAM (Agent Kit Template)
│
└── NEIN → Prüfe Verzeichnisname
├── "lucidlabs-agent-kit" → Wahrscheinlich UPSTREAM
└── Anderer Name → Wahrscheinlich DOWNSTREAM (uninitialisiert)
Context Header anzeigen (IMMER vor Boot Screen):
┌─────────────────────────────────────────────────────────────────────────────┐
│ CONTEXT │
│ ─────── │
│ │
│ Working Directory: /Users/.../projects/invoice-accounting-assistant │
│ Repository Type: DOWNSTREAM │
│ Active Project: invoice-accounting-assistant │
│ PRD: .claude/PRD.md │
│ Upstream: ../../lucidlabs-agent-kit │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
Oder für Upstream (Agent Kit Template):
┌─────────────────────────────────────────────────────────────────────────────┐
│ CONTEXT │
│ ─────── │
│ │
│ Working Directory: /Users/.../lucidlabs-agent-kit │
│ Repository Type: UPSTREAM (Agent Kit Template) │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
WICHTIG: Im Upstream-Modus → Projekt-Auswahl anbieten (siehe Abschnitt 0.0)
Falls Downstream aber kein PROJECT-CONTEXT.md:
┌─────────────────────────────────────────────────────────────────────────────┐
│ CONTEXT │
│ ─────── │
│ │
│ Working Directory: /Users/.../projects/some-project │
│ Repository Type: DOWNSTREAM (uninitialisiert) │
│ │
│ ⚠️ PROJECT-CONTEXT.md fehlt. Erstelle es mit: │
│ → Führe /init-project aus, oder │
│ → Erstelle .claude/PROJECT-CONTEXT.md manuell │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
Nach Context-Erkennung:
Wann: Wenn Claude im lucidlabs-agent-kit Verzeichnis gestartet wird (Upstream).
Workflow:
# Projekte-Ordner ist auf gleicher Ebene wie agent-kit
PROJECTS_DIR="$(dirname "$(pwd)")/projects"
ls -1 "$PROJECTS_DIR" 2>/dev/null | grep -v "^\."
┌─────────────────────────────────────────────────────────────────────────────┐
│ │
│ VERFÜGBARE PROJEKTE │
│ ─────────────────── │
│ │
│ [1] casavi-sandbox-seeder Zuletzt: 28.01.2026 │
│ [2] client-service-reporting Zuletzt: 29.01.2026 │
│ [3] invoice-accounting-assistant Zuletzt: 29.01.2026 │
│ [4] neola Zuletzt: 23.01.2026 │
│ [5] satellite Zuletzt: 29.01.2026 │
│ │
│ ─────────────────────────────────────────────────────────────────────── │
│ │
│ [N] Neues Projekt erstellen │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
An welchem Projekt möchtest du heute arbeiten?
┌─────────────────────────────────────────────────────────────────────────────┐
│ │
│ PROJEKT AUSGEWÄHLT: invoice-accounting-assistant │
│ │
│ Pfad: /Users/.../projects/invoice-accounting-assistant │
│ │
│ ─────────────────────────────────────────────────────────────────────── │
│ │
│ Okay, wollen wir loslegen? │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
KRITISCH: Nach Bestätigung durch den User erfolgt der Handoff automatisch:
a) Working Directory wechseln:
cd /Users/.../projects/[gewähltes-projekt]
b) Handoff-Bestätigung anzeigen:
┌─────────────────────────────────────────────────────────────────────────────┐
│ │
│ ✓ SESSION HANDOFF COMPLETE │
│ │
│ ─────────────────────────────────────────────────────────────────────── │
│ │
│ Neues Working Directory: │
│ /Users/.../projects/invoice-accounting-assistant │
│ │
│ ⚠️ WICHTIG: Ich arbeite ab jetzt NUR in diesem Projekt. │
│ Das Upstream Repository (lucidlabs-agent-kit) wird NICHT verändert. │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
c) Context Header für neues Projekt anzeigen
d) Boot Screen mit verfügbaren Skills
e) Dann: Normaler /prime Flow für Downstream (ab 0.1)
Handoff-Regeln:
| Regel | Beschreibung |
|---|---|
| Kein Session-Neustart | Alles passiert in der gleichen Claude-Session |
| Kein /clear nötig | Working Directory wird gewechselt, Kontext bleibt erhalten |
| Explizite Bestätigung | Handoff wird visuell bestätigt |
| Upstream-Schutz | Nach Handoff: KEINE Änderungen am Agent Kit mehr |
| Projekt-Fokus | Alle weiteren Aktionen betreffen nur das gewählte Projekt |
WANN: Wenn im Upstream-Modus (Agent Kit Template). Zeige nach der Projekt-Liste.
Zweck: Zeige alle pending promote-queue Items aus allen downstream Projekten.
PROJECTS_DIR="$(dirname "$(pwd)")/projects"
echo ""
echo "PROMOTE QUEUE (across all projects)"
echo "────────────────────────────────────"
echo ""
for proj in "$PROJECTS_DIR"/*/; do
QUEUE="$proj/.agents/promote-queue.md"
if [ -f "$QUEUE" ]; then
PENDING=$(grep -c "^### " "$QUEUE" 2>/dev/null || echo "0")
if [ "$PENDING" -gt 0 ]; then
echo "$(basename "$proj"): $PENDING pending items"
grep -A1 "^### " "$QUEUE" | grep -v "^--$" | head -10
echo ""
fi
fi
done
Anzeige:
┌─────────────────────────────────────────────────────────────────────────────┐
│ │
│ PROMOTE QUEUE (across all projects) │
│ ─────────────────────────────────── │
│ │
│ client-service-reporting 3 pending items │
│ → Prime Service Boot v0.1.3 │
│ → Upstream Protection Rule │
│ → Credentials Block v0.1.4 │
│ │
│ cotinga-test-suite 1 pending item │
│ → Cursor Patterns Reference Doc │
│ │
│ ─────────────────────────────────────────────────────────────────────── │
│ │
│ Total: 4 items pending promotion │
│ Use /promote to review and promote items │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
Falls keine pending Items: Block nicht anzeigen.
Rendering-Regeln:
WANN: Im Upstream-Modus. Zeige nach Promote Queue.
Zweck: Zeige wie weit die Downstream-Projekte hinter dem Upstream stehen.
Ausfuehrung:
if [ -f "scripts/drift-detect.sh" ]; then
bash scripts/drift-detect.sh --quiet
fi
Anzeige (wenn Drift vorhanden):
PROJECT SYNC AT BEHIND DAYS SCORE PROMOTE
ll-decision-briefing c406085 0 0 9/9 1
client-service-reporting 7dbf253 8 11 9/9 17
casavi-sandbox-seeder BROKEN --- --- 4/9 0
Rendering-Regeln:
--quiet Modus (kompakte Tabelle ohne Banner)./scripts/audit-sync.sh --dry-runscripts/drift-detect.sh vorhandenWANN: Wenn im Upstream-Modus. Zeige nach Promote Queue.
Zweck: Zeige die nächsten offenen Items aus der Agent Kit Roadmap.
# Show next open items from Agent Kit future-plans
FUTURE=".claude/reference/future-plans.md"
if [ -f "$FUTURE" ]; then
echo "AGENT KIT ROADMAP (next 5)"
echo "──────────────────────────"
grep -E "^- \[ \]" "$FUTURE" | head -5
fi
# Show existing plans
echo ""
echo "ACTIVE PLANS"
echo "────────────"
ls .agents/plans/*.md 2>/dev/null | while read plan; do
echo " → $(basename "$plan" .md)"
done
Anzeige:
┌─────────────────────────────────────────────────────────────────────────────┐
│ │
│ AGENT KIT ROADMAP │
│ ───────────────── │
│ │
│ Next open items: │
│ [ ] PIV-Agent-Teams with hooks │
│ [ ] Model routing definition │
│ [ ] Compound engineering framework │
│ [ ] Marketplace integration (Avery, MCP Skills) │
│ [ ] Portkey integration guide │
│ │
│ Active plans: │
│ → betterauth-org-plugin-and-roles │
│ → integrate-task-system-and-swarm │
│ → migrate-to-skills-standard │
│ → promotion-script │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
WICHTIG: Der Name kommt IMMER aus developer.json. Kein Hardcoding!
TIME_DIR="$HOME/.claude-time"
DEVELOPER_FILE="$TIME_DIR/developer.json"
if [ -f "$DEVELOPER_FILE" ]; then
DEVELOPER_NAME=$(cat "$DEVELOPER_FILE" | jq -r '.name')
fi
Boot Screen Layout (exakt so rendern):
LUCID LABS
───────────────────────────────────────────────────────────────────────────────
█████╗ ██████╗ ███████╗███╗ ██╗████████╗ ██╗ ██╗██╗████████╗
██╔══██╗██╔════╝ ██╔════╝████╗ ██║╚══██╔══╝ ██║ ██╔╝██║╚══██╔══╝
███████║██║ ███╗█████╗ ██╔██╗ ██║ ██║ █████╔╝ ██║ ██║
██╔══██║██║ ██║██╔══╝ ██║╚██╗██║ ██║ ██╔═██╗ ██║ ██║
██║ ██║╚██████╔╝███████╗██║ ╚████║ ██║ ██║ ██╗██║ ██║
╚═╝ ╚═╝ ╚═════╝ ╚══════╝╚═╝ ╚═══╝ ╚═╝ ╚═╝ ╚═╝╚═╝ ╚═╝
{{RANDOM_GREETING}}
───────────────────────────────────────────────────────────────────────────────
{ ────────────────── { * {{RANDOM_AGENT_LOG}} * } ────────────────── }
───────────────────────────────────────────────────────────────────────────────
/agentdocs Explore Agent Kit documentation
/skills List available Agent Kit skills
/clone-skill Clone skills from central repository
/publish-skill Share your skills with the team
/tickets Show assigned Linear tickets
/new-ticket Create a new Linear issue
/todos Open local TODOs and notes
/future View planned features and improvements
/status Show system and session status
/notion-publish Publish markdown to Notion (private)
/session-end End session and persist context
───────────────────────────────────────────────────────────────────────────────
Tip: Type a command or select an option number below.
Danach folgt: Deployed Projects (0.1.0), dann Security Reminders (0.1.0b), dann Tickets/Options-Block (0.3).
IMMER anzeigen nach dem Boot Screen. Lese infrastructure/lucidlabs-hq/registry.json (lokal oder aus upstream) und zeige alle deployed Projekte.
# Registry finden (downstream oder upstream)
REGISTRY=""
if [ -f "infrastructure/lucidlabs-hq/registry.json" ]; then
REGISTRY="infrastructure/lucidlabs-hq/registry.json"
elif [ -f "$(dirname "$(pwd)")/../lucidlabs-agent-kit/infrastructure/lucidlabs-hq/registry.json" ]; then
REGISTRY="$(dirname "$(pwd)")/../lucidlabs-agent-kit/infrastructure/lucidlabs-hq/registry.json"
fi
Deployed Projects anzeigen:
┌─────────────────────────────────────────────────────────────────────────────┐
│ │
│ DEPLOYED PROJECTS (LUCIDLABS-HQ) │
│ ──────────────────────────────── │
│ │
│ cotinga-test-suite │
│ URL: https://cotinga.lucidlabs.de │
│ Convex: https://cts-convex.lucidlabs.de │
│ Repo: https://github.com/lucidlabs-hq/cotinga-test-suite │
│ │
│ client-service-reporting │
│ URL: https://reporting.lucidlabs.de │
│ Convex: https://csr-convex.lucidlabs.de │
│ Repo: https://github.com/lucidlabs-hq/client-service-reporting │
│ │
│ invoice-accounting-assistant Status: pending │
│ URL: https://invoice.lucidlabs.de (not yet deployed) │
│ Repo: https://github.com/lucidlabs-hq/invoice-accounting-assistant │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
Rendering-Regeln:
status: "deployed" oder status: "provisioned" als aktiv anzeigenstatus: "pending" mit Hinweis (not yet deployed) markierenhttps://github.com/<repo> LinkDaten aus registry.json:
{
"projects": [
{
"name": "...",
"url": "https://...",
"convexUrl": "https://...",
"repo": "lucidlabs-hq/...",
"status": "deployed|pending|provisioned"
}
]
}
Prüfe Security-Reminders bei Session-Start.
# Security Reminders laden
REMINDERS_FILE=".claude/reference/security-reminders.json"
if [ ! -f "$REMINDERS_FILE" ]; then
# Versuche upstream
REMINDERS_FILE="$(dirname "$(pwd)")/../lucidlabs-agent-kit/.claude/reference/security-reminders.json"
fi
Logik:
status: "pending"due Datum mit heutigem Datumdue <= heute: Reminder ist OVERDUE (rot/warn)due <= heute + 14 Tage: Reminder ist DUE SOON (gelb/info)Anzeige (nur wenn fällige Reminders existieren):
┌─────────────────────────────────────────────────────────────────────────────┐
│ │
│ SECURITY REMINDERS │
│ ────────────────── │
│ │
│ [OVERDUE] SSH Key Rotation Due: 2026-05-09│
│ Rotate SSH keys for LUCIDLABS-HQ │
│ Ref: .claude/reference/ssh-keys.md │
│ │
│ [DUE SOON] Create dedicated deploy user Due: 2026-08-09│
│ If team has grown: Create 'deploy' user │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
Falls keine Reminders fällig: Block nicht anzeigen (stille Prüfung).
WANN: Bei jedem Session-Start (upstream und downstream). Zeige nach Security Reminders.
Zweck: Proaktiv pruefen ob neue Versionen der Key-Dependencies verfuegbar sind, damit man nicht zu weit zurueckfaellt und bewusst entscheiden kann.
Ausfuehrung:
# Determine package directory
if [ -d "frontend" ]; then
PKG_DIR="frontend"
elif [ -f "package.json" ]; then
PKG_DIR="."
else
PKG_DIR=""
fi
# Run stack monitor (24h cache - won't hit npm on every prime)
if [ -n "$PKG_DIR" ] && [ -f "scripts/stack-monitor.sh" ]; then
bash scripts/stack-monitor.sh --pkg-dir "$PKG_DIR"
elif [ -n "$PKG_DIR" ]; then
# Try upstream script location
UPSTREAM_SCRIPT="$(dirname "$(pwd)")/../lucidlabs-agent-kit/scripts/stack-monitor.sh"
[ -f "$UPSTREAM_SCRIPT" ] && bash "$UPSTREAM_SCRIPT" --pkg-dir "$PKG_DIR"
fi
Anzeige (wenn Updates verfuegbar):
+--------------------------------------------------------------------+
| |
| STACK VERSION MONITOR |
| ----- |
| |
| 1 major, 2 minor, 1 patch update(s) available |
| 2 new (not yet reviewed) |
| |
| DEPENDENCY INSTALLED LATEST TYPE STATUS
| ------------------------------------------------------------------- |
| next 16.0.10 16.1.6 minor NEW |
| react 19.0.0 19.2.1 minor NEW |
| convex 1.28.0 1.31.7 minor PARKED
| zod 3.23.0 4.3.6 MAJOR SKIPPED: uses v3 API
| |
| Review new updates? [y/N] |
| |
+--------------------------------------------------------------------+
Interaktiver Review (wenn User "y" waehlt):
Fuer jedes neue Update:
next 16.0.10 -> 16.1.6 (minor)
Action? [u]pdate / [p]ark / [s]kip with reason / [enter] skip:
| Aktion | Bedeutung | Verhalten bei naechstem /prime |
|---|---|---|
| update | Will updaten | Erinnert: "Run pnpm update next in frontend/" |
| park | Spaeter anschauen | Zeigt wieder als PARKED (bis Version sich aendert) |
| skip + reason | Bewusst auslassen | Zeigt als SKIPPED mit Begruendung, fragt nicht nochmal |
| enter | Erstmal ignorieren | Zeigt wieder als NEW beim naechsten /prime |
Entscheidungen werden gespeichert in .stack-monitor.json:
{
"decisions": {
"zod": {
"parked_version": "4.3.6",
"action": "skipped",
"reason": "uses v3 API throughout codebase, migration too large",
"decided_at": "2026-02-17T22:00:00"
}
},
"last_check": "2026-02-17T22:00:00"
}
Cache-Logik:
.stack-monitor.json ist gitignored (projekt-lokal)Rendering-Regeln:
--quiet und keine neuen UpdatesWANN: Nur bei Downstream-Projekten. Zeige IMMER nach dem Boot Screen.
Zweck: Der Entwickler sieht sofort alle laufenden Services, Ports, Links und Credentials. Keine Suche nötig - sofort produktiv.
Daten sammeln:
PROJECT_DIR="$(pwd)"
PROJECT_NAME=$(basename "$PROJECT_DIR")
# 1. Detect services from directory structure
HAS_FRONTEND=$([ -d "frontend" ] && echo "yes" || echo "no")
HAS_MASTRA=$([ -d "mastra" ] && echo "yes" || echo "no")
HAS_CONVEX=$([ -d "convex" ] && echo "yes" || echo "no")
HAS_N8N=$([ -d "n8n" ] && echo "yes" || echo "no")
# 2. Read ports from package.json
FRONTEND_PORT=$(grep -oE '"dev":\s*".*-p\s+([0-9]+)"' frontend/package.json 2>/dev/null | grep -oE '[0-9]{4}' | tail -1)
MASTRA_PORT=$(grep -oE 'PORT=([0-9]+)' mastra/package.json 2>/dev/null | grep -oE '[0-9]+')
# 3. Read docker-compose ports
if [ -f "docker-compose.dev.yml" ]; then
CONVEX_PORT=$(grep -oE '"([0-9]+):3210"' docker-compose.dev.yml | grep -oE '[0-9]+' | head -1)
DASHBOARD_PORT=$(grep -oE '"([0-9]+):6791"' docker-compose.dev.yml | grep -oE '[0-9]+' | head -1)
fi
# 4. Read .env for credentials info
ENV_FILE=""
[ -f ".env" ] && ENV_FILE=".env"
[ -f ".env.local" ] && ENV_FILE=".env.local"
[ -f "frontend/.env.local" ] && ENV_FILE="frontend/.env.local"
Service Table anzeigen:
┌─────────────────────────────────────────────────────────────────────────────┐
│ │
│ PROJECT SERVICES [project-name] │
│ ──────────────── │
│ │
│ Service Port Link Status │
│ ───────────────── ────── ──────────────────────────── ──────── │
│ Frontend 8070 http://localhost:8070 ○ stopped │
│ Mastra API 8071 http://localhost:8071 ○ stopped │
│ Convex Backend 8072 http://localhost:8072 ○ stopped │
│ Convex Dashboard 8073 http://localhost:8073 ○ stopped │
│ │
│ ─────────────────────────────────────────────────────────────────────── │
│ │
│ CREDENTIALS │
│ ─────────── │
│ │
│ Convex Admin Key: docker exec [project]-convex ./generate_admin_key.sh │
│ Convex Deploy URL: http://localhost:8072 │
│ .env Location: frontend/.env.local │
│ │
│ ─────────────────────────────────────────────────────────────────────── │
│ │
│ STACK MODULES │
│ ───────────── │
│ │
│ [✓] Next.js 16 [✓] Convex [✓] Mastra │
│ [✓] Tailwind CSS 4 [✓] BetterAuth [ ] n8n │
│ [✓] shadcn/ui [ ] Portkey [ ] MinIO │
│ │
│ ─────────────────────────────────────────────────────────────────────── │
│ │
│ ROADMAP (next 5 items from future-plans.md) │
│ ────── │
│ │
│ [ ] Feature A - description │
│ [ ] Feature B - description │
│ [ ] Feature C - description │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
Rendering-Logik:
Service Table:
package.json und docker-compose.dev.yml lesenlsof -i:PORT (● running / ○ stopped)Credentials:
.env Location (welche Datei existiert).env.example existiert, zeige fehlende VariablenStack Modules:
frontend/, mastra/, convex/, n8n/package.json Dependencies: better-auth, @portkey-ai/*, minioRoadmap:
.claude/reference/future-plans.md (zeige nächste 5 offene Items).agents/plans/ und zeige aktive PläneStatus-Check für Services:
# Check if services are running
check_port() {
lsof -i:$1 -sTCP:LISTEN >/dev/null 2>&1 && echo "running" || echo "stopped"
}
FRONTEND_STATUS=$(check_port $FRONTEND_PORT)
MASTRA_STATUS=$(check_port $MASTRA_PORT)
CONVEX_STATUS=$(check_port $CONVEX_PORT)
Stack Module Detection:
# Detect installed modules
HAS_BETTERAUTH=$(grep -q "better-auth" frontend/package.json 2>/dev/null && echo "yes" || echo "no")
HAS_PORTKEY=$(grep -q "portkey" frontend/package.json 2>/dev/null || grep -q "portkey" mastra/package.json 2>/dev/null && echo "yes" || echo "no")
HAS_MINIO=$(grep -q "minio" frontend/package.json 2>/dev/null && echo "yes" || echo "no")
HAS_SHADCN=$([ -f "frontend/components.json" ] && echo "yes" || echo "no")
WICHTIG:
GREETING_POOL (pick 1, use developer.name from JSON):
Welcome back, {{developer.name}}.
Welcome, {{developer.name}}. Session initialized.
Welcome, {{developer.name}}. Ready when you are.
Welcome, {{developer.name}}. Systems over scripts.
Welcome, {{developer.name}}. Let's build something solid.
Welcome, {{developer.name}}. This will compile. Eventually.
AGENT_LOG_POOL (pick 1):
AGENT LOG: THIS STARTED AS A SMALL CHANGE
AGENT LOG: HUMAN IN THE LOOP. FOR NOW.
AGENT LOG: AUTONOMOUS, BUT NOT UNSUPERVISED
AGENT LOG: DESIGNED FOR CLARITY. REALITY MAY VARY
AGENT LOG: OPTIMIZED FOR THINKING, NOT SPEED
AGENT LOG: WORKS ON MY MACHINE
AGENT LOG: TEMPORARY FIX. DO NOT REMOVE
AGENT LOG: TODO – REFACTOR LATER
BOOT SCREEN =
HEADER (static: "LUCID LABS")
LOGO (static: AGENT KIT block letters)
GREETING (random, exactly one line from GREETING_POOL)
DIVIDER
AGENT_LOG (random, exactly one line from AGENT_LOG_POOL, in { } format)
DIVIDER
COMMAND CHEAT SHEET (static)
CONTEXTUAL WORK OPTIONS (dynamic, from Linear/TODOs)
Rules:
Prüfe Developer-Konfiguration:
TIME_DIR="$HOME/.claude-time"
DEVELOPER_FILE="$TIME_DIR/developer.json"
# Prüfe ob Developer-Datei existiert
if [ -f "$DEVELOPER_FILE" ]; then
DEVELOPER_NAME=$(cat "$DEVELOPER_FILE" | grep '"name"' | cut -d'"' -f4)
fi
Falls KEIN Developer konfiguriert (erstes Mal):
{ ──────────────── { * BOOT SEQUENCE * } ──────────────── }
First time setup detected.
lucid labs
────────────────────────────────────────────────────────────────────────────
█████╗ ██████╗ ███████╗███╗ ██╗████████╗ ██╗ ██╗██╗████████╗
██╔══██╗██╔════╝ ██╔════╝████╗ ██║╚══██╔══╝ ██║ ██╔╝██║╚══██╔══╝
███████║██║ ███╗█████╗ ██╔██╗ ██║ ██║ █████╔╝ ██║ ██║
██╔══██║██║ ██║██╔══╝ ██║╚██╗██║ ██║ ██╔═██╗ ██║ ██║
██║ ██║╚██████╔╝███████╗██║ ╚████║ ██║ ██║ ██╗██║ ██║
╚═╝ ╚═╝ ╚═════╝ ╚══════╝╚═╝ ╚═══╝ ╚═╝ ╚═╝ ╚═╝╚═╝ ╚═╝
A modular engineering toolkit for building AI agents
// Setting up your developer profile...
────────────────────────────────────────────────────────────────────────────
Setup-Fragen (AskUserQuestion verwenden):
┌─────────────────────────────────────────────────────────────────────────────────┐
│ │
│ DEVELOPER SETUP │
│ │
│ ───────────────────────────────────────────────────────────────────────── │
│ │
│ Wie heißt du? │
│ > _ │
│ │
│ Deine E-Mail? │
│ > _ │
│ │
│ Dein Kürzel (für Commits, z.B. "adam")? │
│ > _ │
│ │
│ ───────────────────────────────────────────────────────────────────────── │
│ │
│ Linear & Productive.io werden später konfiguriert, │
│ wenn du die entsprechenden Skills nutzt. │
│ │
│ (Gespeichert in ~/.claude-time/developer.json) │
│ │
└─────────────────────────────────────────────────────────────────────────────────┘
Dann speichern in ~/.claude-time/developer.json:
{
"name": "Adam Kassama",
"email": "adam@lucidlabs.de",
"handle": "adam",
"role": "engineer",
"created": "2026-01-28",
"updated": "2026-01-28",
"linear": {
"user_id": "usr_abc123def456",
"display_name": "Adam K.",
"email": "adam@lucidlabs.de",
"default_team_id": "team_xyz789",
"default_team_key": "CUS",
"workspace": "lucid-labs-agents"
},
"productive": {
"person_id": "12345",
"organization_id": "67890",
"email": "adam@lucidlabs.de",
"default_service_id": null,
"default_activity_type_id": "11111",
"rate_card_id": null
},
"time_tracking": {
"total_minutes_all_time": 0,
"first_session": null,
"last_session": null
},
"preferences": {
"boot_humor": true,
"dashboard_style": "compact",
"auto_sync_to_productive": false,
"show_budget_warnings": true
}
}
Felder-Erklärung:
| Feld | Zweck | Wann abfragen | API Source |
|---|---|---|---|
name | Begrüßung, Reports | Erstes Setup | User Input |
email | Identifikation, Matching | Erstes Setup | User Input |
handle | Kurz-ID für Commits | Erstes Setup | User Input |
linear.user_id | API Queries (assignee filter) | Bei erstem /linear | Linear API: viewer.id |
linear.default_team_id | Issue Creation | Bei erstem /linear | Linear API: teams |
linear.default_team_key | Issue Prefix (CUS-xxx) | Bei erstem /linear | Linear API: team.key |
productive.person_id | Time Entry Creation | Bei /productive setup | Productive API: people (match by email) |
productive.organization_id | API Header | Bei /productive setup | Productive API Settings |
productive.default_activity_type_id | Time Entry Type | Bei /productive setup | Productive API: activity_types |
time_tracking.* | Aggregierte Statistiken | Automatisch | Berechnet |
preferences.* | UX-Anpassung | Optional | User Input |
Wie werden die IDs ermittelt?
LINEAR:
1. /linear status → OAuth via MCP
2. Query: viewer { id, email, name }
3. Query: teams { nodes { id, key, name } }
4. Speichere in developer.json
PRODUCTIVE:
1. /productive setup
2. GET /api/v2/people?filter[email]={email}
3. Speichere person_id
4. GET /api/v2/activity_types
5. Zeige Auswahl, speichere default
Falls Developer EXISTIERT (normale Session):
Zeige die Boot Sequence aus 0.1 mit:
Dann direkt weiter zu 0.3 Session-Optionen (Tickets, etc.).
KRITISCH: Before starting a new session, check if a previous session crashed without /session-end.
TIME_DIR="$HOME/.claude-time"
PROJECT_NAME=$(basename "$(pwd)")
CURRENT_SESSION="$TIME_DIR/current-session.txt"
HEARTBEAT_FILE="$TIME_DIR/heartbeat-${PROJECT_NAME}.txt"
HEARTBEAT_PID_FILE="$TIME_DIR/heartbeat-${PROJECT_NAME}.pid"
SESSION_FILE="$TIME_DIR/sessions/$PROJECT_NAME.json"
STALE_THRESHOLD=600 # 10 minutes (2x heartbeat interval)
MAX_SESSION_SECONDS=28800 # 8 hours cap
Recovery-Logik:
Existiert current-session.txt?
├── NEIN → Alles sauber, weiter zu 0.3
└── JA → Vorherige Session wurde nicht beendet!
│
├── Existiert heartbeat-{project}.txt?
│ ├── JA → Lese letzten Heartbeat-Epoch als End-Time
│ └── NEIN → Verwende Start-Epoch + 30min als geschaetzte Dauer
│
├── Berechne Dauer: END_EPOCH - START_EPOCH
│ └── Cap bei MAX_SESSION_SECONDS (8h)
│
├── Speichere recovered Session in sessions/{project}.json:
│ {
│ "date": "2026-02-11",
│ "start": "14:00",
│ "end": "15:25",
│ "duration_minutes": 85,
│ "recovered": true,
│ "recovery_source": "heartbeat" | "estimated",
│ "linear_issue": null,
│ "commits": [],
│ "synced_to_productive": false
│ }
│
├── Zeige Recovery-Banner:
│
│ ┌──────────────────────────────────────────────────────────────────────┐
│ │ SESSION RECOVERY │
│ │ ──────────────── │
│ │ │
│ │ Previous session did not end cleanly. │
│ │ Recovered: ~85 min (via heartbeat) │
│ │ Saved to: ~/.claude-time/sessions/{project}.json │
│ │ │
│ └──────────────────────────────────────────────────────────────────────┘
│
└── Aufraeumen:
rm -f "$CURRENT_SESSION"
rm -f "$HEARTBEAT_FILE"
kill $(cat "$HEARTBEAT_PID_FILE" 2>/dev/null) 2>/dev/null
rm -f "$HEARTBEAT_PID_FILE"
Recovery-Schritte als Bash:
if [ -f "$CURRENT_SESSION" ]; then
# Previous session was not ended cleanly
mkdir -p "$TIME_DIR/sessions"
START_EPOCH=$(grep "^Started:" "$CURRENT_SESSION" | awk '{print $2}')
if [ -f "$HEARTBEAT_FILE" ]; then
END_EPOCH=$(cat "$HEARTBEAT_FILE")
RECOVERY_SOURCE="heartbeat"
else
# No heartbeat found - estimate 30 min session
END_EPOCH=$((START_EPOCH + 1800))
RECOVERY_SOURCE="estimated"
fi
# Calculate duration with 8h cap
DURATION_SECONDS=$((END_EPOCH - START_EPOCH))
if [ "$DURATION_SECONDS" -gt "$MAX_SESSION_SECONDS" ]; then
DURATION_SECONDS=$MAX_SESSION_SECONDS
fi
DURATION_MINUTES=$((DURATION_SECONDS / 60))
# Extract time components for session record
START_TIME=$(date -r "$START_EPOCH" "+%H:%M" 2>/dev/null || date -d "@$START_EPOCH" "+%H:%M" 2>/dev/null)
END_TIME=$(date -r "$END_EPOCH" "+%H:%M" 2>/dev/null || date -d "@$END_EPOCH" "+%H:%M" 2>/dev/null)
SESSION_DATE=$(date -r "$START_EPOCH" "+%Y-%m-%d" 2>/dev/null || date -d "@$START_EPOCH" "+%Y-%m-%d" 2>/dev/null)
# Save recovered session to sessions/{project}.json
# (append to existing sessions array or create new file)
# Show recovery banner
# Clean up stale files
rm -f "$CURRENT_SESSION" "$HEARTBEAT_FILE"
kill "$(cat "$HEARTBEAT_PID_FILE" 2>/dev/null)" 2>/dev/null
rm -f "$HEARTBEAT_PID_FILE"
fi
Nach der Begrüßung zeige die Arbeitsoptionen:
┌─────────────────────────────────────────────────────────────────────────────┐
│ │
│ Woran möchtest du heute arbeiten? │
│ │
│ ─────────────────────────────────────────────────────────────────────── │
│ │
│ MEINE LINEAR TICKETS (dir zugewiesen) │
│ ───────────────────────────────────── │
│ │
│ [1] CUS-42 Login Feature implementieren Delivery 5h 30min │
│ [2] CUS-45 Error Handling verbessern Exploration 1h 15min │
│ [3] CUS-48 API Dokumentation Backlog 0h 00min │
│ │
│ ─────────────────────────────────────────────────────────────────────── │
│ │
│ WEITERE OPTIONEN │
│ ──────────────── │
│ │
│ [4] Future Plans - Geplante Features & Verbesserungen │
│ [5] Lokale TODOs - Deine persönlichen Notizen │
│ [6] Neues Ticket - Neues Linear Issue erstellen │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
Linear Query für "Meine Tickets":
Linear MCP Query:
- Team: [aktuelles Team aus Projekt]
- Assignee: ME (der eingeloggte User)
- Status: NOT Done, NOT Cancelled
- Sort: Status (Delivery first), then Updated
Future Plans laden:
# Show next 10 open items from future-plans.md
cat .claude/reference/future-plans.md 2>/dev/null | grep -E "^- \[ \]" | head -10
Lokale TODOs laden:
# Prüfe ob lokale TODO-Datei existiert
cat ~/.claude-time/todos/[project-name].md 2>/dev/null
WICHTIG: Zeige bei jedem Session-Start das Time Tracking Dashboard!
# Prüfe ob Time Tracking Verzeichnis existiert
TIME_DIR="$HOME/.claude-time"
PROJECT_NAME=$(basename "$(pwd)")
# Erstelle Verzeichnis falls nicht vorhanden
mkdir -p "$TIME_DIR/sessions"
# Lade Session-Daten für dieses Projekt
SESSION_FILE="$TIME_DIR/sessions/$PROJECT_NAME.json"
Zeige das Dashboard im folgenden Format:
╔═══════════════════════════════════════════════════════════════════════════════╗
║ ║
║ SESSION DASHBOARD [project-name] ║
║ ║
╠═══════════════════════════════════════════════════════════════════════════════╣
║ ║
║ AKTIVITÄT (letzte 8 Wochen) ║
║ ─────────────────────────────────────────────────────────────────────── ║
║ ║
║ Dez Jan Jan Jan Jan ║
║ W49 W01 W02 W03 W04 ║
║ ║
║ Mo · · ░ · ▒ · ░ ▒ █ ░ ║
║ Di · · · ░ ▒ ░ ▒ ░ ░ █ ║
║ Mi · · ░ ░ █ ▒ ░ █ ▒ ░ ║
║ Do · · · ▒ ░ █ ▒ ░ █ ▒ ║
║ Fr · · ░ ░ ▒ ░ █ ▒ ░ ░ ║
║ ║
║ Legende: · keine ░ < 1h ▒ 1-3h █ > 3h ║
║ ║
╠═══════════════════════════════════════════════════════════════════════════════╣
║ ║
║ STATISTIK ║
║ ───────── ║
║ ║
║ Gesamtzeit: 24h 30min │ Sessions: 12 ║
║ Diese Woche: 5h 15min │ Aktive Tage: 8/30 ║
║ Heute: 0h 00min │ Ø pro Session: 2h 02min ║
║ ║
║ Längste Session: 4h 30min │ Peak Hour: 10:00-11:00 ║
║ Aktueller Streak: 3 Tage │ Längster Streak: 7 Tage ║
║ ║
╠═══════════════════════════════════════════════════════════════════════════════╣
║ ║
║ LETZTE SESSIONS ║
║ ─────────────── ║
║ ║
║ 27.01.2026 Mo 14:00 - 17:30 3h 30min CUS-42 ✓ synced ║
║ 26.01.2026 So 10:15 - 12:45 2h 30min CUS-41 ✓ synced ║
║ 24.01.2026 Fr 09:00 - 11:00 2h 00min CUS-40 ✓ synced ║
║ 23.01.2026 Do 13:30 - 15:15 1h 45min CUS-39 ○ pending ║
║ ║
╠═══════════════════════════════════════════════════════════════════════════════╣
║ ║
║ KONTINGENT (via Productive.io) [Acme Corp] ║
║ ────────────────────────────── ║
║ ║
║ Budget: 100h Verbraucht: 45h Verbleibend: 55h ║
║ ║
║ ████████████████████████████████████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 45% ║
║ ║
║ Prognose: Bei aktuellem Tempo noch ~4 Wochen bis Budget aufgebraucht ║
║ ║
╚═══════════════════════════════════════════════════════════════════════════════╝
Nach dem Dashboard automatisch neue Session registrieren:
TIME_DIR="$HOME/.claude-time"
PROJECT_NAME=$(basename "$(pwd)")
# Write session file with epoch for crash-safe duration calculation
cat > "$TIME_DIR/current-session.txt" << EOF
Started: $(date +%s)
StartISO: $(date -Iseconds)
Project: $PROJECT_NAME
EOF
# Start heartbeat background process (writes epoch every 5 min)
(while true; do date +%s > "$TIME_DIR/heartbeat-${PROJECT_NAME}.txt"; sleep 300; done) &
HEARTBEAT_PID=$!
echo "$HEARTBEAT_PID" > "$TIME_DIR/heartbeat-${PROJECT_NAME}.pid"
WICHTIG: The heartbeat runs in background and writes the current epoch to heartbeat-{project}.txt every 300 seconds (5 minutes). If the session crashes, the next /prime can use this file to determine the approximate end time of the crashed session.
Falls keine Daten vorhanden (erstes Mal):
╔═══════════════════════════════════════════════════════════════════════════════╗
║ ║
║ SESSION DASHBOARD [project-name] ║
║ ║
╠═══════════════════════════════════════════════════════════════════════════════╣
║ ║
║ Willkommen! Dies ist deine erste Session in diesem Projekt. ║
║ ║
║ Das Time Tracking wird automatisch gestartet. ║
║ Am Ende der Session: /session-end ║
║ ║
║ ──────────────────────────────────────────────────────────────────────── ║
║ ║
║ Für Kontingent-Tracking: ║
║ Konfiguriere ~/.claude-time/project-mapping.json ║
║ ║
╚═══════════════════════════════════════════════════════════════════════════════╝
Query Linear for issues assigned to you or recently updated:
Use Linear MCP to search:
- Team: lucid-labs-agents
- Assignee: me
- Status: Exploration, Decision, Delivery, Review
- Sort: Updated (descending)
Show:
## Linear Status
| ID | Title | Status | Project |
|----|-------|--------|---------|
| ABC-123 | Feature X | Exploration | [Agents] Project |
**New since last session:**
- [list any new issues or comments]
Woran möchtest du arbeiten?
1. [ABC-123] Continue Feature X
2. [New] Start something new
3. [Skip] Just explore codebase
Read PROJECT-STATUS.md to understand:
cat PROJECT-STATUS.md 2>/dev/null || echo "No PROJECT-STATUS.md found"
CLAUDE.md (project rules and conventions).claude/PRD.md (product requirements).claude/reference/ as neededList all tracked files:
git ls-files | head -50
Show directory structure:
find . -type d -not -path '*/node_modules/*' -not -path '*/.git/*' -not -path '*/.next/*' | head -30
If PROJECT-STATUS.md shows an active plan:
# Read the active plan
cat .agents/plans/[active-plan].md 2>/dev/null
Identify:
# Current branch
git branch --show-current
# Recent commits
git log -5 --oneline
# Uncommitted changes
git status --short
# Check Node version
node -v
# Check if dependencies installed
ls frontend/node_modules 2>/dev/null && echo "Dependencies installed" || echo "Run: cd frontend && pnpm install"
## Project Status
**Project:** [Name from PROJECT-STATUS.md or PRD]
**Phase:** [Current phase]
**Branch:** [Git branch]
### Active Plan
[If active plan exists:]
- **Plan:** `.agents/plans/[plan-name].md`
- **Feature:** [Feature being implemented]
- **Progress:** [X/Y tasks completed]
- **Next Task:** [Description of next task]
[If no active plan:]
- No active plan. Ready to start new feature.
### Recent Activity
- [Last 3-5 activities from PROJECT-STATUS.md]
## Codebase Overview
**Tech Stack:**
- Frontend: Next.js [version], React, TypeScript
- Styling: Tailwind CSS v4
- Backend: [Mastra if exists]
- Database: [Convex if configured]
**Structure:**
- `frontend/` - Next.js application
- `mastra/` - AI agent layer (if exists)
- `.claude/` - Documentation & skills
- `.agents/plans/` - Implementation plans
## Ready to Work
### Linear Issues (Active)
[Show issues from Linear in Exploration/Delivery status]
### If Continuing Active Issue:
Issue: [ABC-123] Feature X
Status: [Exploration/Delivery]
Next: [What to do next based on status]
### If Starting New Feature:
1. Create Linear issue first: `/linear create`
2. Then plan: `/plan-feature [feature-name]`
### Available Skills
| Skill | Description |
|-------|-------------|
| `/linear` | Manage Linear issues |
| `/plan-feature` | Create new implementation plan |
| `/execute [plan]` | Execute a plan |
| `/validate` | Run all validation checks |
| `/commit` | Create formatted commit |
| `/session-end` | End session, update Linear |
If there's an active plan with incomplete tasks:
Show the next task clearly:
RESUME POINT
Plan: .agents/plans/[plan-name].md
Task: [Task number and description]
File: [File to work on]
Ready to continue? Say "continue" or ask questions.
Offer to continue:
/plan-featureAt the end, always provide a one-liner:
Quick Resume: [One sentence summary of what to do next]
Examples:
At appropriate stopping points, remind:
Bevor du gehst: /session-end
→ Updates Linear ticket status
→ Adds work summary
→ Ensures clean state for next session
Die folgenden Features sind konzipiert aber noch nicht vollständig implementiert:
/productive setup [customer] - Kunden einrichten
/productive sync - Zeit zu Productive.io syncen
/productive budget - Kontingent-Status anzeigen
Benötigt:
~/.claude-time/todos/[project-name].md
Format:
# Lokale TODOs - customer-portal
## Offen
- [ ] Error Boundary für API Calls
- [ ] Loading States vereinheitlichen
## Ideen
- [ ] Dark Mode Support evaluieren
Skill:
/todo add "Neues TODO"
/todo list
/todo done [id]
Die JSON-Struktur ist definiert, aber das tatsächliche Speichern/Laden muss noch als Code implementiert werden.
Benötigt:
Die Kontingent-Daten kommen von Productive.io und werden im Dashboard angezeigt.
Benötigt: