domain-keyword-detection

name	domain-keyword-detection
description	Detect document domain (14 domains: software/hardware/manufacturing/standards/biotech/business/energy/data_science/automotive/cloud_infrastructure/telecom/iot/medical_devices/security) via keyword matching. Supports multi-domain documents and confidence scoring.

Domain Keyword Detection

Overview

Automatically detect document domain via lightweight keyword matching. Designed for decision extraction, terminology recognition, and content routing to domain-specialized agents.

When to Use This Skill

• Detect document domain(s) for targeted processing
• Activate domain-specific keyword packs for decision extraction
• Route tasks to domain-specialized agents based on detected context
• Validate domain configuration files

Supported Domains

Domain	Description	Example Keywords
software	Software & IT (architecture, tech stacks)	microservices, React, PostgreSQL, MVP
hardware	Power Electronics & Hardware (materials, PD)	nanocrystalline, SiC, power density, liquid cooling
manufacturing	Manufacturing & Supply Chain (processes, SPC)	yield, Cpk, pilot production, rework
standards	Standards & Certification (IEC, IEEE, GB)	IEC, certification, compliance, inter-lab testing
business	Business & Finance (models, metrics, ROI)	ROI, business model, subscription, TCO
biotech	Biotech & Pharma (clinical trials, GMP, FDA)	clinical trial, GMP, target, formulation
energy	Energy & Power Systems (renewable, grid, storage)	photovoltaics, wind power, energy storage, power grid
data_science	Data Science & AI/ML (algorithms, models, training)	machine learning, deep learning, XGBoost, training
automotive	Automotive & EV (powertrain, ADAS, vehicle)	motor, e-drive, BMS, ADAS, autonomous driving
cloud_infrastructure	Cloud & Infrastructure (AWS/GCP/Azure, DevOps)	AWS, Azure, Kubernetes, CI/CD, Terraform
telecom	Telecommunications (5G, network, wireless)	5G, base station, core network, LTE, WiFi
iot	IoT & Embedded Systems (sensors, protocols, edge)	sensors, MQTT, MCU, embedded, RTOS
medical_devices	Medical Devices & Healthcare (FDA 510(k), CE)	FDA, 510(k), ISO13485, clinical trial
security	Cybersecurity & Information Security (penetration testing)	firewall, WAF, penetration testing, MLPS, ISO27001

Supported Commands

Command	Purpose	Exit Codes
detect	Detect document domain(s)	0: success/empty, 1: runtime error, 2: config error
get-keywords	Get keywords for a specific domain	0: success, 1: domain not found
list-domains	List all available domains	0: success
validate	Validate domain YAML files	0: all valid, 1: validation errors

Usage

CLI Commands

# Detect domain(s) from a document
python3 skills/domain-keyword-detection/scripts/domain_detector.py detect \
  --input docs/design.md \
  --threshold 0.3 \
  --output json

# Get keywords for a specific domain
python3 skills/domain-keyword-detection/scripts/domain_detector.py get-keywords --domain hardware

# List all available domains
python3 skills/domain-keyword-detection/scripts/domain_detector.py list-domains

# Validate domain configuration files
python3 skills/domain-keyword-detection/scripts/domain_detector.py validate

Agent Workflow

Decision extraction workflow (ppt-content-planner):

# Step 1: Detect domains from source document
result = run_command(
    "python3 skills/domain-keyword-detection/scripts/domain_detector.py detect",
    args={"input": "docs/design.md", "threshold": 0.3}
)

# Step 2: Use activated packs for decision extraction
activated_packs = result["activated_packs"]  # ["software", "business"]
keywords = {}
for domain in activated_packs:
    keywords[domain] = run_command(
        "python3 skills/domain-keyword-detection/scripts/domain_detector.py get-keywords",
        args={"domain": domain}
    )["keywords"]

# Step 3: Extract key decisions using domain-specific keywords
decisions = extract_decisions(source_doc, keywords)

# Step 4: Report activated packs in content_qa_report.json
report["domain_packs_activated"] = activated_packs

Output Format

Detect Command

{
  "status": "success",
  "detected_domains": ["software"],
  "confidence_scores": {
    "software": 0.85
  },
  "matched_keywords": {
    "software": ["microservices", "React", "PostgreSQL", "MVP", "decision"]
  },
  "activated_packs": ["software"],
  "threshold": 0.3,
  "total_keywords_matched": 12
}

Get-Keywords Command

{
  "status": "success",
  "domain": "hardware",
  "keywords": {
    "decision_verbs": ["decision", "choose", "adopt", "recommend"],
    "technical_terms": ["nanocrystalline", "amorphous", "powder", "SiC", "GaN", "power density"],
    "scope_markers": ["MVP", "Phase 1", "must-have", "optional"],
    "comparison_markers": ["vs", "comparison", "trade-off", "trade-off"],
    "risk_markers": ["risk", "mitigate", "degrade", "monitor"]
  }
}

Domain Configuration (YAML Schema)

Add new domains by creating domains/{domain_name}.yaml:

name: software
description: "Software & IT domain (architecture, tech stacks, algorithms)"
version: 1.0.0
last_updated: 2026-02-06

keywords:
  decision_verbs:
    zh: ["decision", "choose", "adopt", "recommend", "prioritize", "decide"]
    en: ["decision", "chose", "instead of", "vs", "trade-off", "prioritize"]
  
  technical_terms:
    architecture: ["microservices", "monolith", "event-driven", "serverless"]
    tech_stack: ["React", "Vue", "Angular", "PostgreSQL", "MongoDB", "Redis"]
  
  scope_markers:
    zh: ["must-have", "optional", "MVP", "demo", "mass production", "scale-up"]
    en: ["must-have", "optional", "nice-to-have", "MVP", "Phase 1", "pilot"]
  
  comparison_markers:
    zh: ["vs", "comparison", "trade-off", "compromise", "pros", "cons"]
    en: ["vs", "trade-off", "instead of", "pros", "cons", "advantages"]
  
  risk_markers:
    zh: ["to avoid", "mitigate", "degrade", "monitor", "risk", "uncertainty"]
    en: ["to avoid", "mitigate", "fallback", "monitor", "risk", "uncertainty"]

scoring_weights:
  decision_verbs: 1.5
  technical_terms: 1.0
  scope_markers: 1.0
  comparison_markers: 1.0
  risk_markers: 0.8

Confidence Scoring

Algorithm:

Confidence = (matched_unique_keywords / total_domain_keywords) * weight_factor

Weight factors:
- If doc_length < 500 words: penalize (×0.7)
- If matched_keywords > 15: boost (×1.2)

Threshold recommendations:

• 0.2 - Permissive (detect weak domain signals, good for exploration)
• 0.3 - Balanced (default, suitable for most use cases)
• 0.5 - Strict (high-confidence only, reduces false positives)

Return Payloads

Success

status: success
detected_domains: ["software", "business"]
confidence_scores:
  software: 0.85
  business: 0.55
matched_keywords:
  software: ["microservices", "React", "MVP"]
  business: ["ROI", "business model"]
activated_packs: ["software", "business"]
threshold: 0.3
total_keywords_matched: 15

Empty

status: empty
detected_domains: []
confidence_scores: {}
threshold: 0.5
hint: "No domains matched above threshold 0.5. Try lowering threshold or check document content."

Error

status: error
error_message: "Domain file not found: domains/invalid.yaml"
hint: "Check the domain name or create the domain configuration file."

Troubleshooting

Issue	Solution
Confidence scores too low	Lower threshold or enrich domain YAML with more keywords
False positives (wrong domain detected)	Increase threshold or review keyword specificity in domain YAML
Multi-domain confusion	Expected for hybrid documents (e.g., IoT firmware); use confidence scores to prioritize
Domain YAML validation fails	Check YAML syntax, required fields (name, description, keywords), and keyword uniqueness

Limitations & Design Trade-offs

• Lightweight keyword matching (not ML-based): Fast, interpretable, but may miss context-dependent domain signals
• Single-language support per category: Chinese and English keywords mixed; no full i18n support yet
• No semantic understanding: "adopt React" vs "do not adopt React" both match the "React" keyword (use decision extraction for context)
• Fixed threshold: Single global threshold for all domains (future: per-domain thresholds)

Integration

Current users:

• ppt-content-planner — decision extraction with domain-specific keywords

Potential users:

• markdown-writer-specialist — domain-aware glossary and style guide selection
• data-scientist-research-lead — route tasks to domain experts
• cortana — context-aware task routing

Dependencies

• Python 3.8+
• PyYAML (pip3 install pyyaml)

Changelog

• 2026-02-25 — v1.1.0 — Added 8 new domain packs (energy, data_science, automotive, cloud_infrastructure, telecom, iot, medical_devices, security). Total 14 domains now supported.
• 2026-02-06 — v1.0.0 — Initial release with 6 domain packs (software, hardware, manufacturing, standards, business, biotech)