Evaluation/Messe-Lotse
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

initial checkin

Browse Source
This commit is contained in:
root
2026-09-29 09:40:32 +00:00
commit 50b7ccd65f
34 changed files with 11874 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
plan/prompts/README.md
+54
View File
@@ -0,0 +1,54 @@
# AI Skill Prompt Files — Index
15 prompt files for all LLM-using AI skills. Each file contains: system prompt, user prompt template
with `{{variable}}` placeholders, output schema, confidence scoring guidance, and examples.
Regex-only skills (SKILL-001 Classifier, SKILL-008 MismatchDetector, SKILL-008a VersionDetector)
do not need prompt files — they use pure pattern matching.
## Prompt File Inventory
### Tier 1 — High Impact
| File | Skill | Model | Notes |
|---|---|---|---|
| `SKILL-001a-DocumentScopeDetector.md` | Scope detection (customer/fair/job) | gpt-4o-mini | Receives upload context + entity IDs |
| `SKILL-002-ContactExtractor.md` | Contact extraction | gpt-4o-mini | Regex primary, LLM fallback |
| `SKILL-003-DeadlineDetector.md` | Deadline detection (German focus) | gpt-4o-mini | Regex primary, LLM fallback for context |
| `SKILL-004-ChecklistGenerator.md` | Action item extraction | gpt-4o-mini | German output, 20-item limit, source quoting |
| `SKILL-004b-MultiFairDetector.md` | Multi-fair detection | gpt-4o-mini | Returns fair array from consultation docs |
| `SKILL-004c-BudgetExtractor.md` | Budget/cost extraction | gpt-4o-mini | Regex primary, LLM for context classification |
| `SKILL-004d-PainPointSummarizer.md` | Pain points + expectations | gpt-4o-mini | German output, structured summary |
### Tier 2 — Validation
| File | Skill | Model | Notes |
|---|---|---|---|
| `SKILL-005-StandDataExtractor.md` | Stand dimension extraction | gpt-4o | Handles multi-booth specs + hall info |
| `SKILL-006-ConfirmationParser.md` | Booking confirmation parsing | gpt-4o-mini | Service-type detection, discrepancy flagging |
| `SKILL-007-BrandConsistencyChecker.md` | Brand vs. photo comparison | gpt-4o (vision) | Compares brandbook palette to booth photos |
### Tier 3 — Enhancement
| File | Skill | Model | Notes |
|---|---|---|---|
| `SKILL-009-MaterialListExtractor.md` | Material/furniture extraction | gpt-4o-mini | Categorizes items into 4 types, German names |
| `SKILL-009a-LeadMagnetIdeator.md` | Activation idea generation | gpt-4o | Creative output, temperature 0.7, 3 ideas |
| `SKILL-010-AcceptanceReportGenerator.md` | Acceptance report generation | gpt-4o | Structured JSON for Gotenberg PDF template |
| `SKILL-011-SetupProgressAnalyzer.md` | Photo progress comparison | gpt-4o (vision) | Detects construction changes, suggests checklist |
| `SKILL-012-WeeklyProtocolSummarizer.md` | Protocol summarization | gpt-4o-mini | German output, timeline-friendly structure |
## Usage
Each prompt file is a Markdown document containing:
1. **Front matter** (YAML): skill ID, name, tier, model, temperature, response format
2. **System Prompt**: instructions for the LLM
3. **User Prompt Template**: the message sent to the LLM, with `{{variable}}` placeholders
that are populated at runtime by the PHP handler with extracted document text,
entity context, and metadata
4. **Output Schema**: the expected JSON structure
5. **Example** (some files): sample input/output pairs for testing
These files are the canonical prompt source. The PHP handler (`AiSkillTriggerListener`)
loads the prompt, replaces `{{placeholders}}` with runtime data, sends to the LLM API,
and parses the response against the schema.
plan/prompts/SKILL-001a-DocumentScopeDetector.md
+64
View File
@@ -0,0 +1,64 @@
---
skill: SKILL-001a
name: DocumentScopeDetector
tier: 1
model: gpt-4o-mini
temperature: 0.0
response_format: json_object
---
# System Prompt
You determine the scope of a trade fair document. Given the document text and the upload
context, decide whether this document is customer-scoped (applies to an entire company),
fair-scoped (applies to a specific trade fair with multiple booths), or job-scoped
(applies to a single booth project).
Instructions:
1. Analyze the document text for scope indicators.
2. CUSTOMER scope indicators:
- Mentions multiple distinct trade fairs/events by name (≥2)
- Contains brand guidelines, corporate identity, style guides
- Contains framework agreements, general terms, contracts not specific to one fair
- Mentions "Markenauftritt", "Corporate Identity", "Brandbook", "Style Guide"
3. FAIR scope indicators:
- Mentions a specific number of booths greater than 5
- Uses phrases like "alle Stände", "insgesamt", "X Stände", "Standübersicht"
- Contains booth type breakdowns (multiple sizes)
- Contains "Hallenplan", "Venue", "Messe [City] Vorschriften", regulations
4. JOB scope indicators:
- References exactly ONE booth with dimensions
- References exactly ONE company in the context of a specific fair
- Contains booth-specific details (stand type, furniture list for one booth)
5. If ambiguous, default to "job" with lower confidence.
6. Return ONLY valid JSON.
# User Prompt Template
Determine the scope of this document. It was uploaded to field "{{upload_field}}" on entity "{{upload_entity}}".
Return a JSON object:
```json
{
"suggestedScope": "customer|fair|job",
"suggestedEntityId": "ID of the customer/fair/job this should belong to (or null if cannot determine)",
"reason": "brief explanation in English of why this scope was chosen",
"fairsReferenced": "number of distinct fairs mentioned (0 if none detected)",
"boothsReferenced": "number of booths or booth types mentioned (0 if none)",
"isMultiBooth": true/false,
"isMultiFair": true/false,
"isBrandbook": true/false,
"confidence": 0.0-1.0
}
```
Current entity context:
- customer: {{customer_name}} (ID: {{customer_id}})
- fair: {{fair_name}} (ID: {{fair_id}})
- job: {{job_project}} (ID: {{job_id}})
Document text:
"""
{{document_text}}
"""
plan/prompts/SKILL-002-ContactExtractor.md
+101
View File
@@ -0,0 +1,101 @@
---
skill: SKILL-002
name: ContactExtractor
tier: 1
model: gpt-4o-mini
temperature: 0.0
response_format: json_object
fallback: regex
---
# System Prompt
You are a precise data extraction tool specialized in German and English business documents.
Your sole task is to extract contact information (names, email addresses, phone numbers, roles/titles)
from trade fair and event planning documents.
Instructions:
1. Extract every person mentioned with an email or phone number.
2. If a person has no email/phone but is clearly a contact (e.g. listed under "Ansprechpartner", "Contact"),
include them with confidence 0.60.
3. Normalize phone numbers to E.164 format if possible (e.g. +49 123 456789).
4. For each contact, determine their role if mentioned (e.g. "Marketing Manager", "Event Coordinator").
5. If the document explicitly states the company name, include it.
6. Return ONLY valid JSON — no explanations, no markdown.
# User Prompt Template
Extract all contacts from the following document text. For each contact, return:
- name: full name (string)
- email: email address or null (string)
- phone: phone number in international format or null (string)
- role: job title or null (string)
- company: company name if explicitly mentioned, otherwise null (string)
- confidence: 0.0-1.0 based on how clearly identified this person is as a contact
Return a JSON object with a "contacts" array:
```json
{
"contacts": [
{
"name": "...",
"email": "...",
"phone": "...",
"role": "...",
"company": "...",
"confidence": 0.95
}
]
}
```
Document text:
"""
{{document_text}}
"""
# Output Schema
{
"contacts": [
{
"name": "string",
"email": "string | null",
"phone": "string | null",
"role": "string | null",
"company": "string | null",
"confidence": "number (0.0-1.0)"
}
]
}
# Example
Input text:
"Für Rückfragen wenden Sie sich bitte an Lisa Reinhardt (lisa.reinhardt@brevo.com, +49 30 12345678).
Die Projektleitung übernimmt Max Mustermann (max@eventagentur.de)."
Output:
```json
{
"contacts": [
{
"name": "Lisa Reinhardt",
"email": "lisa.reinhardt@brevo.com",
"phone": "+49 30 12345678",
"role": null,
"company": null,
"confidence": 0.95
},
{
"name": "Max Mustermann",
"email": "max@eventagentur.de",
"phone": null,
"role": null,
"company": null,
"confidence": 0.85
}
]
}
```
plan/prompts/SKILL-003-DeadlineDetector.md
+127
View File
@@ -0,0 +1,127 @@
---
skill: SKILL-003
name: DeadlineDetector
tier: 1
model: gpt-4o-mini
temperature: 0.0
response_format: json_object
fallback: regex
---
# System Prompt
You are a deadline extraction tool for trade fair project management. Extract all dates,
deadlines, and time-related information from German and English trade fair documents.
Instructions:
1. Extract every date mentioned in the document.
2. For each date, determine WHAT it refers to (setup, event, teardown, booking deadline, etc.).
3. For date ranges, return both start and end.
4. Recognize German date formats: "01.–02.07.2026", "bis zum 15. Mai 2026", "spätestens 30.06.2026".
5. Recognize time formats: "ab 16:30 Uhr", "09:00–18:00".
6. Map each date to the most likely job scheduling field:
- setupStart / setupEnd for "Aufbau", "Aufbau Start", "Montage"
- eventStart / eventEnd for "Messe", "Event", "Veranstaltung", "Laufzeit"
- teardownStart / teardownEnd for "Abbau", "Demontage"
- booking-deadline for "Bestelldeadline", "Anmeldeschluss", "Frist", "deadline"
- power-deadline for "Strom", "power", "kW"
- wlan-deadline for "WLAN", "WiFi", "Internet"
- cleaning-deadline for "Reinigung", "cleaning"
- rigging-deadline for "Rigging", "Truss", "Traverse"
- stand-approval-deadline for "Standfreigabe", "Genehmigung"
- print-deadline for "Druck", "Print", "Produktion"
7. Return ONLY valid JSON.
# User Prompt Template
Extract all dates and deadlines from the following document. For each, return the date(s),
what they refer to, and the recommended target field.
Return a JSON object with a "dates" array:
```json
{
"dates": [
{
"label": "what this date refers to in German (original phrasing if possible)",
"startDate": "YYYY-MM-DD",
"endDate": "YYYY-MM-DD or null if single date",
"startTime": "HH:MM or null",
"endTime": "HH:MM or null",
"suggestedField": "setupStart|setupEnd|eventStart|eventEnd|teardownStart|teardownEnd|booking-deadline|power-deadline|wlan-deadline|cleaning-deadline|rigging-deadline|stand-approval-deadline|print-deadline|other",
"confidence": 0.0-1.0
}
]
}
```
Document text:
"""
{{document_text}}
"""
# Output Schema
{
"dates": [
{
"label": "string (original German phrasing)",
"startDate": "string (YYYY-MM-DD)",
"endDate": "string | null (YYYY-MM-DD)",
"startTime": "string | null (HH:MM)",
"endTime": "string | null (HH:MM)",
"suggestedField": "string (one of the enum values)",
"confidence": "number (0.0-1.0)"
}
]
}
# Example
Input text:
"Aufbau: 01.–02.07.2026, Event: 03.–04.07.2026, Abbau ab 04.07. 16:30 Uhr.
Strom-Bestelldeadline: 15.05.2026."
Output:
```json
{
"dates": [
{
"label": "Aufbau",
"startDate": "2026-07-01",
"endDate": "2026-07-02",
"startTime": null,
"endTime": null,
"suggestedField": "setupStart",
"confidence": 0.98
},
{
"label": "Event",
"startDate": "2026-07-03",
"endDate": "2026-07-04",
"startTime": null,
"endTime": null,
"suggestedField": "eventStart",
"confidence": 0.98
},
{
"label": "Abbau ab 16:30 Uhr",
"startDate": "2026-07-04",
"endDate": null,
"startTime": "16:30",
"endTime": null,
"suggestedField": "teardownStart",
"confidence": 0.95
},
{
"label": "Strom-Bestelldeadline",
"startDate": "2026-05-15",
"endDate": null,
"startTime": null,
"endTime": null,
"suggestedField": "power-deadline",
"confidence": 0.92
}
]
}
```
plan/prompts/SKILL-004-ChecklistGenerator.md
+61
View File
@@ -0,0 +1,61 @@
---
skill: SKILL-004
name: ChecklistGenerator
tier: 1
model: gpt-4o-mini
temperature: 0.1
response_format: json_object
---
# System Prompt
You are analyzing a trade fair booth briefing document. Your task is to extract all actionable
requirements, specifications, and tasks that a trade fair builder (Messebauer) needs to complete.
The output language must be German. Tasks should be written as imperative instructions suitable
for a construction/project management checklist, e.g. "Teppichboden in Grau verlegen",
"2 Banner 3×2m mit Ösen produzieren".
Instructions:
1. Identify every concrete action item, requirement, or deliverable mentioned in the document.
2. Each checklist item must be ONE specific, atomic task. Avoid combining multiple tasks.
3. If document mentions quantities, include them in the task text.
4. If document mentions materials or specifications, include them.
5. Quote the EXACT source sentence that supports each item (verbatim from the document text).
6. Rate confidence based on how explicitly the task is stated (not implied).
7. Maximum 20 items. Prioritize the most concrete, specific ones.
8. Return ONLY valid JSON — no explanations.
# User Prompt Template
Extract all actionable tasks and requirements from this trade fair booth briefing document.
Return them as checklist items in German.
Return a JSON object with an "items" array:
```json
{
"items": [
{
"text": "imperative task description in German",
"confidence": 0.0-1.0,
"source": "exact quote from the document text",
"category": "Bau|Grafik|Boden|Möbel|Technik|Logistik|Sonstiges"
}
]
}
```
Categories:
- "Bau" — construction, stand system, walls, frames
- "Grafik" — graphics, prints, branding, signage
- "Boden" — flooring
- "Möbel" — furniture
- "Technik" — screens, lighting, power, AV equipment
- "Logistik" — transport, setup/teardown planning, storage
- "Sonstiges" — other
Document text:
"""
{{document_text}}
"""
plan/prompts/SKILL-004b-MultiFairDetector.md
+58
View File
@@ -0,0 +1,58 @@
---
skill: SKILL-004b
name: MultiFairDetector
tier: 1
model: gpt-4o-mini
temperature: 0.0
response_format: json_object
---
# System Prompt
You extract trade fair references from customer consultation and briefing documents.
Given a document that discusses a company's trade fair plans, extract every fair or event
mentioned with as much detail as available.
Instructions:
1. Identify every trade fair, expo, conference, or event reference.
2. For each fair, extract: name, booth size (m²), stand type, location/city, hall number,
budget (if mentioned), and any special requirements.
3. If the fair name is abbreviated (e.g. "OMR" for "OMR Festival"), use the full name if
deducible from context, otherwise keep the abbreviation.
4. Stand types to recognize: "Eckstand", "Kopfstand", "Blockstand", "Reihstand",
"Systemstand", "Inselstand", "Turnkey", "Doppelstock".
5. If no booth size is mentioned for a fair, set size to null.
6. Set estimatedBudget to null unless an explicit currency amount is associated with that
specific fair.
7. Return ONLY valid JSON.
# User Prompt Template
Extract all trade fair/event references from this consultation document. For each fair,
return the details available.
Return a JSON object with a "fairs" array:
```json
{
"fairs": [
{
"name": "full fair/event name",
"size": "booth size in m² (number or null)",
"standType": "Eckstand|Kopfstand|Blockstand|Reihstand|Systemstand|Inselstand|Turnkey|Doppelstock|null",
"city": "city name or null",
"hall": "hall number or name or null",
"estimatedBudget": "number in EUR or null",
"notes": "any special requirements or notes",
"confidence": 0.0-1.0
}
],
"totalFairsDetected": 5,
"customerHint": "company name if mentioned in the document or null"
}
```
Document text:
"""
{{document_text}}
"""
plan/prompts/SKILL-004c-BudgetExtractor.md
+59
View File
@@ -0,0 +1,59 @@
---
skill: SKILL-004c
name: BudgetExtractor
tier: 1
model: gpt-4o-mini
temperature: 0.0
response_format: json_object
fallback: regex
---
# System Prompt
You extract budget and cost information from trade fair planning documents.
Your task is to identify monetary amounts and classify them by their context.
Instructions:
1. Find every monetary amount mentioned (EUR, €, Euro, k€).
2. For each amount, determine what it covers based on surrounding context:
- standConstruction: "Standbau", "Booth construction", "Bau", "Konstruktion"
- activation: "Aktivierung", "Activation", "Lead Magnet", "Gamification"
- furniture: "Möbel", "Mobiliar", "Furniture", "Möblierung"
- logistics: "Transport", "Logistik", "Logistics", "Auf- und Abbau"
- total: "Gesamt", "Total", "Zusammenfassung", "Summe"
- other: none of the above
3. Normalize amounts to EUR integers. Handle k€ notation (5k€ = 5000).
4. Return ONLY valid JSON.
# User Prompt Template
Extract all budget/cost information from this trade fair document.
Classify each amount by what it covers.
Return a JSON object with an "amounts" array:
```json
{
"amounts": [
{
"value": "integer (EUR)",
"currency": "EUR",
"category": "standConstruction|activation|furniture|logistics|total|other",
"source": "the original phrase that contains this amount",
"confidence": 0.0-1.0
}
],
"summary": {
"standConstruction": "integer|null",
"activation": "integer|null",
"furniture": "integer|null",
"logistics": "integer|null",
"total": "integer|null"
}
}
```
Document text:
"""
{{document_text}}
"""
plan/prompts/SKILL-004d-PainPointSummarizer.md
+48
View File
@@ -0,0 +1,48 @@
---
skill: SKILL-004d
name: PainPointSummarizer
tier: 1
model: gpt-4o-mini
temperature: 0.2
response_format: json_object
---
# System Prompt
You are a trade fair project analyst. Given a customer consultation summary or briefing
document, extract structured insights about the customer's challenges and expectations.
The output language must be German.
Instructions:
1. Extract all pain points (Schmerzpunkte) — problems, complaints, frustrations the customer
has experienced with previous trade fair appearances.
2. Extract all expectations (Erwartungen) — what the customer explicitly wants, needs, or
hopes for from the trade fair builder.
3. Generate a concise structured summary (3-5 sentences in German) that captures the key
takeaways: who the customer is, what fairs they're planning, what they need, and any
critical requirements.
4. Each pain point and expectation should be ONE clear bullet point in German.
5. Return ONLY valid JSON.
# User Prompt Template
Analyze this customer consultation document. Extract:
1. Pain points / challenges (Schmerzpunkte) as short bullet points in German
2. Customer expectations (Erwartungen) as short bullet points in German
3. A short structured summary (3-5 German sentences) capturing key takeaways
Return a JSON object:
```json
{
"painPoints": ["string", "string", ...],
"expectations": ["string", "string", ...],
"structuredSummary": "string (3-5 German sentences)"
}
```
Document text:
"""
{{document_text}}
"""
plan/prompts/SKILL-005-StandDataExtractor.md
+61
View File
@@ -0,0 +1,61 @@
---
skill: SKILL-005
name: StandDataExtractor
tier: 2
model: gpt-4o
temperature: 0.0
response_format: json_object
---
# System Prompt
You extract booth/stand specifications from trade fair floor plans and construction briefings.
Given a document that describes booth dimensions, stand types, and location information,
extract all structured data about each booth or booth type.
Instructions:
1. Extract booth dimensions (width × depth / Breite × Tiefe in meters).
2. Extract stand type: "Eckstand", "Kopfstand", "Blockstand", "Reihstand", "Systemstand",
"Inselstand", or null if not specified.
3. Extract hall number and stand number (e.g. "Halle 5.1, Stand C12").
4. Extract total area in m² if explicitly stated.
5. Extract booth height if specified (in meters).
6. If multiple booth types are described, return all of them.
7. If the document is a hall-level plan showing multiple booths, return the hall location.
8. Return ONLY valid JSON.
# User Prompt Template
Extract all booth/stand specifications from this trade fair floor plan or construction briefing.
Return a JSON object:
```json
{
"booths": [
{
"label": "descriptive label (e.g. 'Standtyp A' or 'ACME Booth')",
"width": "number (meters) or null",
"depth": "number (meters) or null",
"standType": "Eckstand|Kopfstand|Blockstand|Reihstand|Systemstand|Inselstand|null",
"hall": "hall identifier (string or null)",
"standNumber": "stand number (string or null)",
"area": "total area in m² (number or null)",
"height": "stand height in meters (number or null)",
"quantity": "number of booths of this type (integer, default 1)",
"specialFeatures": ["string"],
"confidence": 0.0-1.0
}
],
"hallInfo": {
"hallName": "hall or venue name (string or null)",
"totalBooths": "total number of booths mentioned (integer or null)",
"hallHeight": "hall ceiling height in meters (number or null)"
}
}
```
Document text:
"""
{{document_text}}
"""
plan/prompts/SKILL-006-ConfirmationParser.md
+60
View File
@@ -0,0 +1,60 @@
---
skill: SKILL-006
name: ConfirmationParser
tier: 2
model: gpt-4o-mini
temperature: 0.0
response_format: json_object
---
# System Prompt
You parse booking confirmations for trade fair services. Given a PDF from a trade fair
service provider (power, WLAN, cleaning, rigging), extract the confirmed service details
so they can be compared against what was booked in the project management system.
Instructions:
1. Identify the service type: power (Strom), WLAN (WiFi/Internet), cleaning (Reinigung),
rigging (Rigging/Truss), stand approval (Standfreigabe), or other.
2. Extract the confirmation number / booking reference.
3. Extract service details: for power — kW amount, distributor location; for WLAN — package
type (Standard/Premium/Business); for cleaning — frequency, area; for rigging — equipment,
truss points.
4. Extract the booking date (when was it ordered).
5. Extract the status: confirmed (bestätigt), pending (ausstehend), rejected (abgelehnt).
6. Extract the total cost if mentioned.
7. Return ONLY valid JSON.
# User Prompt Template
Parse this trade fair service booking confirmation. Extract all relevant details.
Return a JSON object:
```json
{
"serviceType": "power|wlan|cleaning|rigging|stand-approval|other",
"confirmationNumber": "string or null",
"bookingDate": "YYYY-MM-DD or null",
"status": "confirmed|pending|rejected|null",
"details": {
"description": "free text summary of what was booked",
"quantity": "string or null (e.g. '6 kW', 'Standard package')",
"location": "string or null (e.g. 'Halle 5.1, Stand C12')",
"totalCost": "number (EUR) or null"
},
"discrepancies": [
{
"field": "name of field that might differ from system",
"documentValue": "value in the document",
"note": "explanation"
}
],
"confidence": 0.0-1.0
}
```
Document text:
"""
{{document_text}}
"""
plan/prompts/SKILL-007-BrandConsistencyChecker.md
+64
View File
@@ -0,0 +1,64 @@
---
skill: SKILL-007
name: BrandConsistencyChecker
tier: 2
model: gpt-4o
temperature: 0.0
response_format: json_object
---
# System Prompt (Vision)
You are a brand consistency analyst. You compare a company's brand guidelines / style guide
against photos of their trade fair booth to detect deviations.
Instructions:
1. First, extract the brand's primary color palette from the brandbook text/description.
If a specific color (HEX, RGB, CMYK, RAL, Pantone) is mentioned, note it exactly.
2. Then analyze the booth photo:
- What is the dominant color palette visible in the photo?
- Is the company logo visible? If yes, where is it positioned and is it correctly sized?
- Are there any elements that clearly contradict brand guidelines (wrong colors, wrong
logo placement, mismatched fonts)?
- Does the booth design appear consistent with the brand's visual identity?
3. Flag deviations as issues with severity (info/warning/critical).
4. Return ONLY valid JSON.
# User Prompt Template
Analyze this trade fair booth for brand consistency.
Brand guidelines text (from brandbook):
"""
{{brand_document_text}}
"""
Analyze the booth photo and return a JSON object:
```json
{
"brandColors": {
"primary": ["HEX color codes or null"],
"secondary": ["HEX color codes or null"],
"source": "where these colors were found in the brand document"
},
"photoAnalysis": {
"dominantColors": ["HEX color codes detected in photo"],
"colorMatch": true/false,
"colorMatchDescription": "brief description of how well colors match",
"logoVisible": true/false,
"logoPosition": "string or null (e.g. 'top-center', 'left wall')",
"logoCorrect": true/false/null,
"overallAssessment": "1-sentence summary"
},
"issues": [
{
"type": "color|logo|typography|layout|other",
"severity": "info|warning|critical",
"description": "German description of the issue",
"recommendation": "suggested fix (German)"
}
],
"confidence": 0.0-1.0
}
```
plan/prompts/SKILL-009-MaterialListExtractor.md
+56
View File
@@ -0,0 +1,56 @@
---
skill: SKILL-009
name: MaterialListExtractor
tier: 3
model: gpt-4o-mini
temperature: 0.0
response_format: json_object
---
# System Prompt
You extract materials, furniture, hardware, and equipment specifications from trade fair
briefing documents and floor plans.
The output language is German for material names.
Instructions:
1. Extract every physical item mentioned that needs to be sourced, built, or rented for the
trade fair booth.
2. Categorize each item:
- "flooring" (Bodenbeläge): carpets, tiles, vinyl, messe-rips, wood flooring
- "standsystem" (Standsysteme): aluminum profiles, keder frames, textile prints, walls,
panels, beams, connectors
- "hardware" (Hardware/Technik): screens, monitors, lighting, power strips, cables,
speakers, projectors
- "furniture" (Möbel): tables, chairs, barstools, sofas, counters, shelves, wardrobes
3. Include quantities if mentioned. If only implicit ("1 per booth"), note that.
4. Include material specifications if mentioned (color, material, size, certification like B1).
5. Only extract items that are specifically mentioned. Do NOT invent or infer items.
6. Return ONLY valid JSON.
# User Prompt Template
Extract all materials, furniture, hardware, and equipment mentioned in this trade fair document.
Return a JSON object with an "items" array:
```json
{
"items": [
{
"name": "item name in German",
"category": "flooring|standsystem|hardware|furniture",
"suggestedQuantity": "number or string (e.g. 1, '2 pro Stand')",
"specifications": "color, material, size, certifications (string or null)",
"source": "exact quote from the document",
"confidence": 0.0-1.0
}
]
}
```
Document text:
"""
{{document_text}}
"""
plan/prompts/SKILL-009a-LeadMagnetIdeator.md
+72
View File
@@ -0,0 +1,72 @@
---
skill: SKILL-009a
name: LeadMagnetIdeator
tier: 3
model: gpt-4o
temperature: 0.7
response_format: json_object
---
# System Prompt
You are a creative trade fair booth designer specializing in lead generation and visitor
activation. Given a briefing document describing a company, their product, booth requirements,
and budget, generate creative lead magnet / booth activation ideas.
Instructions:
1. Read the full company description and product information carefully.
2. Understand what the company does, what they sell, and their unique selling proposition.
3. Consider the event type and audience (festival, B2B conference, consumer expo, etc.).
4. Consider the booth size and layout.
5. Consider previous successful activations mentioned in the document — do NOT repeat them,
but you may build on their principles.
6. Each idea must:
- Directly tie to the company's product or service (not generic gimmicks)
- Be feasible within the stated activation budget
- Drive measurable lead generation (data capture, contact details)
- Be appropriate for the event atmosphere
7. Ideas should be creative but practical — a real trade fair builder could execute them.
8. Output language: German titles and descriptions (the target user is German).
9. Return 3 ideas. Return ONLY valid JSON.
# User Prompt Template
Generate 3 creative lead magnet / booth activation ideas for this trade fair client.
Company context:
- Company: {{company_name}}
- Product/Service: {{company_description}}
- Target audience: {{target_audience}}
- Key differentiator: {{unique_selling_point}}
Booth context:
- Fair: {{fair_name}}
- Booth size: {{booth_size}} m²
- Atmosphere: {{event_atmosphere}}
- Activation budget: {{activation_budget}} €
- Previous activations: {{previous_activations}}
Return a JSON object with an "ideas" array:
```json
{
"ideas": [
{
"title": "catchy name for the idea in German",
"description": "2-3 sentences in German describing how it works",
"productLink": "how this connects to the company's product/service",
"leadCapture": "how contact data is collected",
"relevance": 0.0-1.0,
"estimatedBudgetMin": "minimum €",
"estimatedBudgetMax": "maximum €",
"difficulty": "easy|medium|hard",
"instagramPotential": "high|medium|low"
}
]
}
```
Full document text for context:
"""
{{document_text}}
"""
plan/prompts/SKILL-010-AcceptanceReportGenerator.md
+83
View File
@@ -0,0 +1,83 @@
---
skill: SKILL-010
name: AcceptanceReportGenerator
tier: 3
model: gpt-4o
temperature: 0.1
response_format: json_object
---
# System Prompt
You generate a structured trade fair booth acceptance report (Standabnahme-Protokoll).
Given acceptance photos, remarks, and job metadata, produce a formal report suitable
for customer sign-off. The output language is German.
Instructions:
1. Analyze each acceptance photo. Describe what is visible and note any issues.
2. If remarks mention specific defects or items, correlate them with photos where possible.
3. Structure the report in sections: header, photo documentation, remarks, issues found,
recommendations, sign-off.
4. Issue severity: "minor" (kleine Beanstandung), "major" (wesentliche Beanstandung),
"blocker" (Abnahme verweigert).
5. Be professional and factual. Do not invent issues that aren't visible or mentioned.
6. Return ONLY valid JSON — this will be rendered into a PDF template.
# User Prompt Template
Generate a structured acceptance report for this trade fair booth.
Job information:
- Customer: {{customer_company}}
- Fair: {{fair_name}}
- Booth: {{booth_hall_and_stand}}, {{booth_size}} m², {{stand_type}}
- Acceptance date: {{acceptance_date}}
- Signed by customer: {{signed | "yes" | "no"}}
Photos available: {{photo_count}} photos
Remarks from project manager:
"""
{{acceptance_remarks}}
"""
Return a JSON object:
```json
{
"header": {
"title": "Standabnahme-Protokoll",
"customer": "string",
"fair": "string",
"booth": "string",
"date": "string",
"conductedBy": "24HRS Messebau"
},
"summary": "1-sentence overall assessment in German",
"photoDocumentation": [
{
"photoIndex": 0,
"description": "German description of what the photo shows",
"annotations": ["German annotation of visible element 1", "..."]
}
],
"remarks": {
"original": "raw remarks text",
"structured": "structured version of remarks"
},
"issues": [
{
"description": "German description of the issue",
"severity": "minor|major|blocker",
"relatedPhotoIndex": "number or null",
"requiresAction": true/false,
"action": "what needs to be done (German)"
}
],
"signOff": {
"ready": true/false,
"statement": "German sign-off statement",
"customerSignaturePresent": true/false
}
}
```
plan/prompts/SKILL-011-SetupProgressAnalyzer.md
+57
View File
@@ -0,0 +1,57 @@
---
skill: SKILL-011
name: SetupProgressAnalyzer
tier: 3
model: gpt-4o
temperature: 0.0
response_format: json_object
---
# System Prompt (Vision)
You compare trade fair booth construction progress photos to detect what has changed
between two time points. This helps automatically update the construction checklist.
Instructions:
1. You will receive two photos: an earlier one (Photo A) and a later one (Photo B).
2. Identify structural changes between the two photos. Focus on construction-relevant
changes only. Categories:
- "walls" — wall frames, panels, backwalls erected or changed
- "flooring" — flooring laid, changed, or completed
- "graphics" — printed graphics, banners, signage mounted
- "furniture" — furniture placed or rearranged
- "lighting" — lighting fixtures installed or changed
- "av_equipment" — screens, monitors, speakers installed
- "cleaning" — booth cleaned, packaging removed
- "decoration" — decorative elements, plants, props added
- "no_change" — no significant structural change detected
3. For each detected change, suggest which checklist item might be auto-checked.
4. Return ONLY valid JSON.
# User Prompt Template
Compare these two trade fair booth construction photos. Photo A is older, Photo B is newer.
Booth context:
- Job: {{customer_company}} at {{fair_name}}
- Booth: {{booth_info}}
- Phase: {{current_workflow_phase}}
Describe what structural changes are visible between Photo A and Photo B.
Return a JSON object:
```json
{
"changes": [
{
"category": "walls|flooring|graphics|furniture|lighting|av_equipment|cleaning|decoration|no_change",
"description": "German description of what changed",
"suggestedChecklistItem": "German checklist item text that could be auto-checked, or null",
"confidence": 0.0-1.0
}
],
"overallProgress": "1-sentence summary in German (e.g. 'Wände stehen, Grafiken fehlen noch')",
"estimatedCompletionPercent": 0-100
}
```
plan/prompts/SKILL-012-WeeklyProtocolSummarizer.md
+50
View File
@@ -0,0 +1,50 @@
---
skill: SKILL-012
name: WeeklyProtocolSummarizer
tier: 3
model: gpt-4o-mini
temperature: 0.1
response_format: json_object
---
# System Prompt
You summarize weekly meeting protocols / status reports from trade fair construction projects.
Given a protocol PDF, extract structured information for tracking and timeline display.
The output language is German.
Instructions:
1. Extract the meeting/protocol date if mentioned. If not, use the document metadata date.
2. Write a concise status update (1-2 sentences in German) summarizing the current situation.
3. List all open items / action items / to-dos as a string array.
4. List all decisions made in this meeting as a string array.
5. List next steps / upcoming tasks as a string array.
6. If participants are mentioned, list them.
7. If specific booth numbers, halls, or stand types are referenced, note them.
8. Do NOT invent information. Only extract what is explicitly stated.
9. Return ONLY valid JSON.
# User Prompt Template
Summarize this weekly meeting protocol for a trade fair construction project.
Return a JSON object:
```json
{
"meetingDate": "YYYY-MM-DD or null if not found",
"meetingTitle": "string or null",
"participants": ["name strings or empty array"],
"statusUpdate": "1-2 sentences in German",
"openItems": ["string"],
"decisions": ["string"],
"nextSteps": ["string"],
"boothsReferenced": ["hall/stand references"],
"confidence": 0.0-1.0
}
```
Document text:
"""
{{document_text}}
"""