Generates Texas Property Code-compliant demand letters for habitability and Fair Housing Act discrimination claims. Two distinct templates with proper statutory citations.

Habitability Template (TPC 92.056)

Required info: Landlord name/address, property address, unit number, lease start/end dates, monthly rent, specific conditions (mold, plumbing, pest, HVAC, structural), dates noticed landlord, repair request dates, landlord responses.

Output structure:

• Header with dates and addresses
• Property identification and lease terms
• Statutory basis (TPC 92.052 duty to repair, 92.056 landlord liability)
• Specific conditions with dates and documentation references
• Timeline of repair requests and landlord failures
• Demanded remedies: repair completion, rent reduction, actual damages
• Statutory deadline: 7-day cure period per TPC 92.056(b)(1)
• Treble damages warning if not resolved (DTPA 17.50)

Discrimination Template (FHA 42 USC 3604)

Required info: Protected class, specific discriminatory acts, dates, witnesses, comparative treatment evidence.

Sending instructions: Certified mail with return receipt requested. Keep copies. Send to registered agent if corporate landlord.

Follow-up timeline: 7 days for habitability, 30 days for discrimination before filing.

Launches 4 sub-agents in parallel, each searching a different Qdrant collection. Synthesizes results with cross-references.

Architecture

# What happens when you run /legal-search "DTPA treble damages"
Agent 1 -> qdrant: texas_caselaw    (state court opinions)
Agent 2 -> qdrant: texas_primary_law (statutes, TRCP)
Agent 3 -> qdrant: bia_precedents    (if immigration query)
Agent 4 -> qdrant: uscis_policy      (if immigration query)

Flags

Output

Returns a synthesized summary with: top 5 most relevant results per collection, case citations with holdings, statutory references with section numbers, cross-references between findings.

Systematic evaluation of case strength before filing. Analyzes elements, defenses, evidence gaps, and strategic recommendations.

Output includes: Element-by-element analysis, evidence strength rating (strong/moderate/weak), identified defenses with counter-strategies, statute of limitations check, venue and jurisdiction analysis, damages calculation estimate, recommended next steps.

Breaks down a statute into its elements, maps available facts to each element, and identifies gaps.

Method: Identify statutory text, parse elements (each "and"/"or" creates a branch), map evidence to each element, flag elements with insufficient evidence, cite relevant case law interpreting ambiguous elements.

Generates a complete filing package for Texas District Court. Produces 4 documents and detailed filing instructions.

Generated Documents

1. Original Petition -- Complete with cause number placeholder, parties section, jurisdiction/venue, factual allegations organized by claim, causes of action with statutory elements, damages prayer, jury demand, verification
2. Civil Case Information Sheet -- Required by Texas courts: case type codes, monetary range, discovery level, related cases
3. Summons -- Directed to each defendant with proper return-of-service language
4. Filing Instructions -- Step-by-step guide for the courthouse or eFileTexas

Petition Template Structure

CAUSE NO. _______________

PLAINTIFF,                    IN THE DISTRICT COURT
v.
DEFENDANT(S),                 _____ JUDICIAL DISTRICT

                              DALLAS COUNTY, TEXAS

PLAINTIFF'S ORIGINAL PETITION

I. DISCOVERY CONTROL PLAN
II. PARTIES
III. JURISDICTION AND VENUE
IV. FACTUAL BACKGROUND
V. CAUSES OF ACTION
  A. DTPA Violations (Tex. Bus. & Com. Code 17.46)
  B. Breach of Warranty of Habitability (TPC 92.052)
  C. Fair Housing Act Violation (42 USC 3604)
VI. DAMAGES
VII. CONDITIONS PRECEDENT
VIII. JURY DEMAND
IX. PRAYER FOR RELIEF

Filing Methods

Service of Process

Methods: Personal service by constable/process server ($75-150), certified mail with return receipt, service by publication (last resort, requires court order), waiver of service (defendant agrees to accept).

Post-filing steps: File return of service, request default if no answer within 20 days (Monday after service), calendar all deadlines.

The largest single skill in the legal toolkit. Generates TRCP-compliant motions for every common situation in Texas civil litigation. Each motion includes proper caption, certificate of service, proposed order, and specific TRCP rule citations.

All 17 Motion Types

Each motion includes: Proper caption with cause number, parties, court. Factual background specific to the motion type. Legal authority with Texas case citations. Specific prayer for relief. Certificate of service. Proposed order for judge's signature. Notice of hearing (if applicable).

# Usage
/motion-drafter                  # Interactive -- asks which motion type
/motion-drafter "compel"         # Direct -- generates discovery compel motion
/motion-drafter "summary"        # Generates traditional summary judgment

Generates a full discovery package tailored to a DTPA landlord-tenant case. All questions are specifically crafted for short-term rental (STR) evidence and habitability claims.

Contents

STR-Specific Questions (examples)

# Interrogatory targeting STR misrepresentation
"State whether the subject property was listed on any short-term rental
platform (including but not limited to Airbnb, VRBO, Booking.com) during
the 24 months preceding the filing of this suit, and for each listing
provide: (a) the platform name, (b) listing URL, (c) dates active,
(d) nightly rate range, (e) description of amenities advertised."

# RFA targeting habitability knowledge
"Admit that prior to Plaintiff's occupancy, Defendant received at least
one complaint regarding [specific condition] at the subject property."

Builds a 10-section trial notebook structure for Texas District Court proceedings.

Sections

1. Case Summary -- One-page overview: parties, claims, key facts, theme, damages sought
2. Cast of Characters -- All witnesses, parties, and key figures with roles and contact info
3. Chronology -- Master timeline of all events, color-coded by category
4. Legal Elements -- Each claim broken into elements with supporting evidence mapped to each
5. Exhibit List -- Numbered exhibits with description, authentication method, foundation witness, admission status
6. Witness Outlines -- Direct and cross-examination outlines per witness, with key questions and anticipated objections
7. Opening Statement -- Thematic outline with evidence preview and narrative structure
8. Closing Argument -- Element-by-element argument tying evidence to jury charge
9. Jury Charge -- Proposed instructions, definitions, and special issues
10. Trial Day Checklist -- Physical items to bring, courtroom procedures, emergency contacts

Witness Outline Template

WITNESS: [Name]
ROLE: [Plaintiff / Defendant / Expert / Fact]
DIRECT EXAMINATION:
  Q1: State your name for the record.
  Q2: [Foundation questions - who, what, where]
  Q3: [Key testimony questions - mapped to elements]
  [Exhibit introduction: "I'm showing you what's been marked as Exhibit __"]

ANTICIPATED CROSS / IMPEACHMENT:
  - [Likely attack on credibility]
  - [Prior inconsistent statement: depo page __, line __]

REHABILITATION:
  - [Redirect questions to cure damage from cross]

The main search script. Runs across both EOSM3 and EOSmini, combining filesystem tools with AI-powered extraction.

# Usage
sylvan-scan "DTPA violation evidence"           # Standard scan (phases 1-4)
sylvan-scan "Hello Landing STR income" --all    # All 6 phases
sylvan-scan "mold inspection report" --deep     # Phases 1-4 + LLM extraction
sylvan-scan "rent payments" --ocr               # Include OCR phase for scanned docs

Phases

Output: Writes results to /tmp/sylvan-scan-results.json with per-phase results, file paths, match snippets, and confidence scores. Returns a summary to stdout.

Deeper than sylvan-scan. Starts with ripgrep candidates, enriches with Qdrant semantic search, then runs qwen2.5:32b for AI extraction from top results.

# Usage
sylvan-deep "evidence of landlord knowledge of conditions"
sylvan-deep "Hello Landing revenue from unit" --json

Steps

1. Candidate Selection: rg finds files containing any query terms. Ranks by hit density.
2. Semantic Enrichment: Top 20 candidates + Qdrant semantic search results merged via RRF score fusion.
3. LLM Extraction: qwen2.5:32b reads top 10 documents and extracts specific answers to the query.

Model: Uses qwen2.5:32b (18GB VRAM) for extraction -- best structured output model available locally.

Specialized extraction using local LLMs. Each type uses a different prompt template optimized for that data type.

# Usage
sylvan-extract entities "lease.pdf"       # Names, orgs, addresses, amounts
sylvan-extract dates "communications/"    # All dates with context
sylvan-extract violations "inspection.md" # DTPA/habitability violations
sylvan-extract comms "emails/"            # Sender, recipient, date, subject, key points
sylvan-extract money "financial-records/" # Dollar amounts with parties and context
sylvan-extract promises "lease.pdf"       # Promises, warranties, representations
sylvan-extract all "04-STR-Evidence/"     # Run all 7 types

Extraction Types

Note: Uses deepseek-r1:32b for legal reasoning tasks (violations, promises) and qwen2.5:32b for structured data extraction.

Builds an entity relationship graph from case documents using nawalkhan/legal-llm (or falls back to qwen2.5:32b).

# Usage
sylvan-graph "04-STR-Evidence/"            # Build graph from directory
sylvan-graph "lease.pdf" --update          # Add to existing graph
sylvan-graph --query "Hello Landing"       # Query existing graph

Output Sections

1. Entity List: All extracted entities with types and frequencies
2. Relationship Map: Entity-to-entity connections (e.g., "CAF Management" --manages--> "Lofts at Sylvan Thirty")
3. Evidence Chains: Document-to-claim connections showing which files support which legal elements

Stats: Current graph has 1,943 entities extracted from 277 documents, with cross-referenced relationships.

Combines BM25 keyword scores with Qdrant dense vector scores using Reciprocal Rank Fusion (RRF).

# Usage
sylvan-hybrid "treble damages DTPA"         # Hybrid search
sylvan-hybrid "lease breach" --alpha 0.7    # More weight on semantic
sylvan-hybrid "rent amount" --alpha 0.3     # More weight on keyword

Alpha parameter: 0.0 = pure keyword (BM25), 1.0 = pure semantic (Qdrant vectors). Default: 0.6 (tuned for legal queries).

Deduplication: Uses AWK-based dedup to merge results from both engines, keeping highest-scoring entry for each document.

# Modes
sylvan-search "query"               # Quick (fd + rg only)
sylvan-search "query" --deep        # Deep (adds Qdrant + LLM)
sylvan-search "query" --fuzzy       # Fuzzy matching (approximate)

# Filters
sylvan-search "query" -t pdf        # File type filter
sylvan-search "query" -d 30         # Modified in last 30 days
sylvan-search "query" -m 5          # Max 5 results
sylvan-search "query" -p "04-STR"   # Path filter (specific folder)
sylvan-search "query" -l            # List mode (filenames only)

sylvan-duplicates -- MD5 Hash Dedup

# Find duplicate files by content hash
sylvan-duplicates                   # Scan all case folders
sylvan-duplicates "04-STR-Evidence" # Scan specific folder
sylvan-duplicates --delete          # Interactive deletion of dupes

sylvan-history -- Query Tracking

# Track search history for replay
sylvan-history                      # Show recent queries
sylvan-history --replay 5           # Replay query #5
sylvan-history --stats              # Search frequency stats

sylvan-recent -- Recently Modified

# Find recently changed files
sylvan-recent                       # Last 24 hours
sylvan-recent 7                     # Last 7 days
sylvan-recent 30 --type pdf         # PDFs modified in last 30 days

The master investigation skill. Deploys 7 sub-agents across 2 batches using all 56 tools. Three speed modes.

Speed Modes

Batch 1 (Fast Scans -- parallel)

1. Username Agent: Sherlock + Blackbird + NexFil -- fast username existence checks
2. Email Agent: Holehe + h8mail + Buster -- email platform registration + breach data
3. Infrastructure Agent: WHOIS + DNS dig + dnstwist -- domain registration and network data

Batch 2 (Deep Scans -- informed by batch 1)

4. Deep Username Agent: Maigret (3,000+ sites) + Social Analyzer -- finds accounts batch 1 missed
5. Domain Agent: TheHarvester + Subfinder + Amass -- full subdomain and email enumeration
6. Comprehensive Agent: SpiderFoot (200+ modules) -- automated multi-source recon
7. Correlation Agent: Synthesizes all findings, identifies connections between data points

# Usage
/intel username:johndoe email:john@example.com    # Standard mode
/intel domain:example.com --quick                  # Quick mode
/intel "John Doe" --deep                           # Deep mode, all data types

Tiers

Phases

1. Corporate Records: Business filings, incorporation, registered agent, officers/directors
2. Digital Footprint: Domain age, tech stack, web presence, social media profiles
3. Financial Signals: Revenue estimates, funding rounds, job postings (growth signals)
4. Leadership Profiles: Person-intel on key executives and board members
5. Litigation History: Court records, lawsuits, regulatory actions
6. Competitive Landscape: Similar companies, market positioning, differentiation
7. Risk Assessment: Sanctions checks, adverse media, supply chain risks

1. WHOIS: Registration data, registrar, dates, privacy status
2. DNS Records: A, AAAA, MX, TXT, NS, SOA, CNAME via dig
3. Subdomains: subfinder + amass + sublist3r -- passive and active enumeration
4. Tech Stack: webcheck -- server software, CMS, frameworks, analytics
5. SSL/TLS: Certificate details, chain, expiry, transparency logs
6. Port Scan: nmap service scan on discovered hosts
7. Typosquatting: dnstwist -- registered similar domains
8. Historical: waybackurls -- past URLs and archived pages
9. Email Harvesting: theharvester -- emails associated with domain
10. Document Metadata: metagoofil -- metadata from publicly hosted documents

/social-scan (5 Phases)

1. Platform enumeration (sherlock + blackbird)
2. Profile data extraction (instaloader + snscrape)
3. Content analysis (posting patterns, topics, sentiment)
4. Connection mapping (followers, following, interactions)
5. Cross-platform correlation (same person across platforms)

/breach-check (4 Phases)

1. Email breach lookup (h8mail -- HIBP, BreachDirectory, IntelX)
2. Password hash analysis (cracked/uncracked status)
3. Data dump search (paste sites, dark web snippets via IntelX)
4. Risk assessment (severity, recency, exposed data types)

/image-osint

Tools: exiftool (EXIF metadata: GPS, camera, timestamp), greverse (Google reverse image search), deepface (facial analysis). Use for: Verifying image authenticity, finding original sources, extracting location data.

/data-brokers

Checks 14 people-search sites for personal data exposure. Generates opt-out request templates for each site. Sites include: Spokeo, WhitePages, BeenVerified, TruePeopleSearch, FastPeopleSearch, Radaris, Intelius, PeopleFinder, USSearch, ThatsThem, Pipl, ZabaSearch, Classmates, MyLife.

/public-records (5 Record Types)

1. Property records (county assessor data, DCAD for Dallas)
2. Court records (civil, criminal, federal via PACER)
3. Business filings (Secretary of State, UCC filings)
4. Voter registration (state voter rolls)
5. Professional licenses (state licensing boards)

/opsec -- Operational Security

Reference document for investigator OPSEC. Covers: VPN usage, browser isolation, account separation, metadata stripping, search engine privacy, social engineering awareness. Not a tool -- a reference guide to consult before starting sensitive investigations.

/osint-analyst (4 Modes)

• Mentor mode: Suggests techniques and tools for a given investigation goal
• Analysis mode: Helps connect dots between findings from different sources
• Technique library: Reference of 50+ OSINT techniques organized by data type
• Hypothesis testing: Helps form and test investigative hypotheses

/case-setup -- Investigation Setup

Creates a structured investigation directory:

investigation-[target]-[date]/
  01-scope/          # Target definition, investigation plan
  02-raw-data/       # Unprocessed tool outputs
  03-processed/      # Cleaned and organized findings
  04-analysis/       # Synthesis, connection maps, timelines
  05-evidence/       # Screenshots, exports, verifiable proof
  06-reports/        # Final deliverables
  README.md          # Investigation log and methodology

Multi-stage pipeline that processes tax documents through EOSmini's HuggingFace OCR models.

Pipeline Stages

1. Upload: Document sent to EOSmini via SSH (scp to /tmp/tax-processing/) ~2s
2. Classify: classify_document identifies the form type (W-2, 1099, receipt, etc.) ~3s
3. OCR: ocr_form extracts fields using form-specific templates ~8s
4. Validate: Cross-checks extracted values (e.g., do box totals match?) ~1s
5. Categorize: Maps extracted data to Schedule C categories using vendor database ~2s
6. Output: Structured JSON + human-readable summary ~1s

Supported Documents

Two tool groups: 11 tax-data lookup tools and 28 tax-calc computation tools.

Tax Data Tools (11)

Tax Calculation Tools (28)

Three-tier vendor categorization with fuzzy matching and anomaly detection.

Lookup Pipeline

1. Exact Match: Check vendor name against 108+ entries in ~/.claude-money/vendor-categories.json
2. Fuzzy Match: If no exact match, use rapidfuzz with threshold >= 85%. Handles misspellings, abbreviations, partial names.
3. Web Search Fallback: For unknown vendors, search to determine business type, then categorize and add to database with --add flag.

Database Structure

{
  "STRIPE": {
    "category": "Payment Processing",
    "schedule_c_line": 10,
    "line_name": "Commissions and fees",
    "type": "EXPENSE",
    "confidence": 100
  },
  "CANVA": {
    "category": "Marketing/Design",
    "schedule_c_line": 8,
    "line_name": "Advertising",
    "type": "EXPENSE",
    "confidence": 100
  }
}

Anomaly Detection

Uses Z-score analysis to flag unusual transactions:

• Amount anomaly: Transaction amount > 2 standard deviations from vendor average
• Frequency anomaly: Unusual number of transactions with a vendor in a period
• Category mismatch: Transaction description suggests different category than vendor default

Profession-Specific Checklists

Pre-built checklists for common self-employment types (freelance developer, catering, rideshare, real estate agent) with expected vendor types and deduction categories.

# Usage
/smart-categorize "STRIPE"                    # Look up known vendor
/smart-categorize "ACME WIDGET CO" --add 18   # Add unknown vendor to Line 18
python3 ~/.claude-money/scripts/categorize_vendor.py "STRIPE"  # Direct CLI

Searches email and documents for vendor contracts, subscription renewals, and payment terms.

• Searches Gmail for vendor-related emails (invoices, contracts, renewals)
• Tracks renewal dates and alerts on upcoming renewals
• Cross-references with bank transactions to verify charges match contracts
• Flags subscriptions without matching contracts