Claude Cowork Plugins: The Complete Implementation Guide

1. Download Claude Desktop for macOS from https://claude.com/download
2. Install the application
3. Sign in with your Claude account (Pro, Max, Team, or Enterprise)

1. Open Claude Desktop
2. Look for the mode selector showing "Chat" and "Cowork" tabs
3. Click the "Cowork" tab to switch to Tasks mode

1. Cowork will prompt you to grant folder access
2. Best practice: Create a dedicated work folder first
   mkdir ~/cowork-workspace
3. Click "Choose Folder" and select ~/cowork-workspace
4. Grant permissions when prompted

Organize the files in this folder by type and create a summary report.

my-plugin/
├── .claude-plugin/
│   └── plugin.json          # Required: Plugin manifest
├── commands/                 # Slash commands (Markdown files)
│   ├── extract-rfp.md
│   └── analyze-contracts.md
├── agents/                   # Sub-agent definitions
│   ├── requirement-extractor.md
│   ├── risk-analyzer.md
│   └── financial-parser.md
├── skills/                   # Auto-invoked instructions
│   ├── legal-taxonomy/
│   │   └── SKILL.md
│   └── financial-ratios/
│       └── SKILL.md
├── hooks/
│   └── hooks.json           # Lifecycle automation
├── .mcp.json                # External tool integrations
└── scripts/                 # Helper utilities
    └── format-output.sh

---
description: Extract all requirements from RFP documents
arguments: <folder-path>
---
# RFP Requirements Extraction
Analyze all RFP documents in the specified folder and extract:
1. **Functional Requirements**: Features, capabilities, specifications
2. **Technical Requirements**: Platforms, integrations, security standards
3. **Compliance Requirements**: Certifications, regulatory mandates
4. **Evaluation Criteria**: Scoring rubrics, decision factors, weighting
For each requirement:
- Assign unique ID (REQ-001, REQ-002...)
- Categorize as Mandatory or Optional
- Note RFP page number for citation
Output: Excel workbook with requirements matrix, coverage tracking
formulas, and risk flags.
Arguments provided by user: $ARGUMENTS

---
name: requirement-extractor
description: Extract structured requirements from RFP sections
model: sonnet-4.5
tools: [Read, Grep]
permissions:
  file_access: read_only
---
You are a requirements extraction specialist. Your role is to:
1. Read RFP sections assigned to you
2. Identify all requirements (functional, technical, compliance)
3. Extract exact requirement text with page citations
4. Classify each as mandatory vs. optional
5. Return structured JSON to orchestrator
DO NOT:
- Make assumptions about unstated requirements
- Editorialize or paraphrase requirement language
- Access files outside your assigned scope
Return findings in this format:
{
  "requirements": [
    {
      "id": "REQ-001",
      "category": "Functional",
      "text": "System must support SSO via SAML 2.0",
      "mandatory": true,
      "page": 23
    }
  ]
}

{
  "mcpServers": {
    "salesforce": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-salesforce"],
      "env": {
        "SALESFORCE_USERNAME": "${SALESFORCE_USERNAME}",
        "SALESFORCE_TOKEN": "${SALESFORCE_TOKEN}"
      }
    },
    "google-sheets": {
      "command": "${CLAUDE_PLUGIN_ROOT}/servers/sheets-server.js",
      "args": ["--credentials", "${GOOGLE_CREDENTIALS_PATH}"],
      "cwd": "${CLAUDE_PLUGIN_ROOT}"
    }
  }
}

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Write|Edit",
        "hooks": [
          {
            "type": "command",
            "command": "${CLAUDE_PLUGIN_ROOT}/scripts/format-excel.sh",
            "description": "Auto-format Excel files after creation"
          }
        ]
      }
    ]
  }
}

{
  "decision": "approve",
  "reason": "File formatted successfully",
  "systemMessage": "Added conditional formatting and formulas"
}

---
name: legal-clause-taxonomy
description: Classification system for contract clauses.
  Use when analyzing legal documents.
---
# Legal Clause Taxonomy
When analyzing contracts, identify these clause categories:
## Risk Clauses
- **Indemnification**: Who compensates whom for damages
  (look for "indemnify", "hold harmless")
- **Limitation of Liability**: Caps on damages
  (look for "aggregate liability shall not exceed")
- **Warranty Disclaimers**: Limitation of guarantees
  (look for "AS IS", "WITHOUT WARRANTY")
## Operational Clauses
- **Termination Rights**: How contract can be ended
  (look for "terminate for convenience", notice periods)
- **Auto-Renewal**: Automatic contract extension
  (look for "evergreen", "automatically renew")
- **Governing Law**: Which jurisdiction applies
  (look for "governed by laws of")
## Intellectual Property
- **IP Ownership**: Who owns work product
  (look for "work for hire", "assigns all rights")
- **License Grants**: Permission to use IP
  (look for "non-exclusive license", scope, territory)
Flag as **HIGH RISK** if you find:
- Unlimited liability
- Perpetual licenses with broad scope
- One-sided indemnification (we indemnify them, they don't indemnify us)

{
  "name": "rfp-assassin",
  "version": "1.2.0",
  "description": "Extract 100% of requirements from RFP documents",
  "author": {
    "name": "Your Team",
    "email": "team@company.com",
    "url": "https://github.com/yourteam"
  },
  "homepage": "https://docs.yourcompany.com/rfp-assassin",
  "repository": "https://github.com/yourteam/rfp-assassin",
  "license": "MIT",
  "keywords": ["rfp", "sales", "requirements", "proposals"],
  "commands": ["./custom/commands/special.md"],
  "agents": "./custom/agents/",
  "skills": "./custom/skills/",
  "hooks": "./config/hooks.json",
  "mcpServers": "./mcp-config.json",
  "lspServers": "./.lsp.json",
  "dependencies": {
    "plugins": ["base-analytics"],
    "mcpServers": ["salesforce"]
  },
  "configuration": {
    "env": {
      "DEFAULT_OUTPUT_FORMAT": "excel",
      "MAX_RFP_PAGES": "200"
    }
  }
}

1. Open Claude Desktop → Navigate to Cowork tab
2. Click "Plugins" in left sidebar
3. Browse library → Find plugin you need
4. Click plugin → Review description and components
5. Click "Install" button
6. Plugin downloads and activates automatically

1. Click "Upload plugin" button
2. Select plugin folder or ZIP file from your computer
3. Claude validates manifest and installs

1. View installed plugin in Plugins sidebar
2. Click "Customize" button (upper right corner)
3. Claude automatically prompts: "How would you like to customize this plugin?"
4. Describe changes:
   - "Add integration with our internal CRM at https://crm.company.com"
   - "Change output format to Google Sheets instead of Excel"
   - "Include our brand voice guidelines when drafting content"
5. Click "Let's go" → Claude modifies plugin configuration
6. Test customized plugin with sample task

mkdir -p resume-screener/.claude-plugin
cd resume-screener

{
  "name": "resume-screener",
  "version": "1.0.0",
  "description": "Screen and rank job candidates from resume folders",
  "author": {
    "name": "Your HR Team"
  },
  "keywords": ["hr", "recruiting", "resumes", "ats"]
}

---
description: Screen resumes against job requirements and rank candidates
arguments: <job-description-file> <resume-folder>
---
# Resume Screening Workflow
You are an expert ATS (Applicant Tracking System) performing
initial resume screening.
## Inputs
- **Job Description**: $ARGUMENTS (first argument - path to JD file)
- **Resume Folder**: $ARGUMENTS (second argument - folder of resumes)
## Process
### Step 1: Extract Job Requirements
Read the job description and extract:
- **Required Skills**: Must-have technical and soft skills
- **Required Experience**: Minimum years in relevant domain
- **Required Education**: Degree level and fields
- **Nice-to-Have**: Preferred qualifications
- **Location**: Remote/hybrid/onsite requirements
### Step 2: Parse Resumes (Parallel Sub-agents)
For each resume in the folder:
1. Extract candidate information:
   - Name, contact info, location
   - Work experience (titles, companies, dates, achievements)
   - Skills (technical and soft)
   - Education (degrees, institutions, years)
2. Calculate match score:
   - Required skills match: % of must-haves present
   - Experience: Does candidate meet minimum years?
   - Education: Does candidate have required degree?
   - Nice-to-have bonus: Extra points for preferred qualifications
3. Flag red flags:
   - Employment gaps >12 months
   - Job hopping (<1 year tenures)
   - Skill mismatches
### Step 3: Rank and Output
Generate Excel spreadsheet with:
- **Columns**: Rank, Name, Match %, Years Experience, Key Skills,
  Education, Location, Interview Recommendation
- **Sorting**: Ranked by match score (highest first)
- **Conditional Formatting**:
  - Green: Match >80% (Strong Yes)
  - Yellow: Match 60-80% (Yes)
  - Orange: Match 40-60% (Maybe)
  - Red: Match <40% (No)
- **Formulas**: Auto-calculate match percentage based on skills present
Save output as: candidate-rankings.xlsx

---
name: resume-parser
description: Extract structured data from individual resumes
model: sonnet-4.5
tools: [Read]
permissions:
  file_access: read_only
---
You are a resume parser. Your task is to extract structured
information from a single resume and return it in JSON format.
Extract:
1. **Contact**: name, email, phone, LinkedIn, location
2. **Experience**: List of jobs with title, company, start_date,
   end_date, key_achievements
3. **Skills**: Array of technical and soft skills mentioned
4. **Education**: Degrees with institution, field, graduation_year
Return only valid JSON, no commentary:
{
  "contact": {
    "name": "Jane Smith",
    "email": "jane@email.com",
    "phone": "555-1234",
    "linkedin": "linkedin.com/in/janesmith",
    "location": "San Francisco, CA"
  },
  "experience": [
    {
      "title": "Senior Product Manager",
      "company": "Tech Corp",
      "start_date": "2020-01",
      "end_date": "2025-12",
      "years": 5,
      "achievements": ["Launched 3 products", "Grew ARR 200%"]
    }
  ],
  "skills": ["Product Strategy", "SQL", "Figma", "Stakeholder Management"],
  "education": [
    {
      "degree": "MBA",
      "institution": "Stanford GSB",
      "field": "Business Administration",
      "graduation_year": 2019
    }
  ]
}

---
name: ats-criteria
description: ATS screening best practices and bias mitigation.
  Use when screening candidates.
---
# ATS Screening Criteria
## Best Practices
### Skills Matching
- Exact match: "Python" in JD → look for "Python" in resume
- Synonym awareness: "JavaScript" = "JS" = "ECMAScript"
- Avoid keyword stuffing detection: Flag resumes with >50 skill mentions
### Experience Calculation
- Count only relevant experience (not total career length)
- Startup experience: Weight 1.5x (broader responsibility)
- Internships: Count as 0.5x full-time experience
### Education Evaluation
- Bootcamp graduates: Acceptable for technical roles if strong portfolio
- Self-taught: Evaluate based on project work, not credentials
- Degree field mismatch: Don't auto-reject (career changers can be excellent)
## Bias Mitigation
**Remove from consideration** (to avoid discrimination):
- Age (graduation dates, years of experience can indicate age)
- Gender (inferred from name)
- Race/ethnicity
- Parental status (resume gaps often indicate caregiving)
**Focus on**:
- Demonstrated skills (projects, achievements, results)
- Growth trajectory (promotions, increasing responsibility)
- Cultural adds (diverse perspectives) not "cultural fit"
**Red Flag False Positives**:
- Employment gaps: May indicate caregiving, health, education → ask, don't auto-reject
- Job hopping: Tech industry norm is 2-3 years → recalibrate expectations
- Non-traditional paths: Bootcamp grads, self-taught often outperform
## Legal Compliance
- EEOC guidelines: No discrimination based on protected classes
- Blind screening: Remove names/photos before ranking if possible
- Document decisions: Explain why candidates were rejected (defensible criteria)

# Navigate to parent directory containing your plugin
cd ..
# Test plugin before installing
claude --plugin-dir ./resume-screener
# In Claude session, try your command
/resume-screener:screen-resumes job-description.txt /path/to/resumes/

claude plugin install ./resume-screener --scope user

claude plugin install ./resume-screener --scope project

1. Open Claude Desktop → Cowork tab
2. Click "Plugins" → "Upload plugin"
3. Select resume-screener folder
4. Confirm installation

Plugin Name: rfp-assassin

Components:
- Slash Command: /rfp:extract
- Sub-agents:
    - requirement-extractor.md (parallel per section)
    - criteria-parser.md (evaluation rubrics)
    - timeline-mapper.md (milestones, deadlines)
- Skills:
    - rfp-taxonomy/ (SOW structures, compliance matrices)
    - requirement-classification/ (mandatory vs optional logic)
- Hook: PostToolUse → auto-format Excel with conditional formatting
- MCP: Salesforce (pull existing opportunity data)

/rfp:extract /RFPs/Acme-Corp-RFP-2026/

{
  "section": "Technical Requirements",
  "requirements": [
    {
      "id": "REQ-001",
      "category": "Integration",
      "text": "System must integrate with SAP ERP via REST API",
      "mandatory": true,
      "page": 23,
      "keywords": ["SAP", "REST API", "integration"]
    },
    {
      "id": "REQ-002",
      "category": "Security",
      "text": "All data must be encrypted at rest using AES-256",
      "mandatory": true,
      "page": 24,
      "keywords": ["encryption", "AES-256", "data at rest"]
    }
  ]
}

Plugin Name: 10k-analyst

Components:
- Slash Command: /finance:analyze-10k
- Sub-agents (one per fiscal year):
    - 10k-extractor-FY2021.md
    - 10k-extractor-FY2022.md
    - 10k-extractor-FY2023.md
    - 10k-extractor-FY2024.md
    - 10k-extractor-FY2025.md
- Skills:
    - gaap-standards/ (accounting rule interpretations)
    - financial-ratios/ (calculation formulas)
    - sec-filing-structure/ (where to find specific data)
- MCP: SEC EDGAR API (fetch filings automatically)

/finance:analyze-10k TSLA 2021-2025

Sub-agent: 10k-extractor-FY2023

Task: Extract all financial data from Tesla FY2023 10-K

Extract:
1. Income Statement (Revenue, COGS, Operating Expenses, Net Income)
2. Balance Sheet (Assets, Liabilities, Equity)
3. Cash Flow (Operating, Investing, Financing)
4. Key Metrics (mentioned in MD&A)
5. Risk Factors (new risks added this year)
6. Segment Data (Automotive, Energy, Services)

Return structured JSON with exact figures and page citations.

KEY THEMES FROM MANAGEMENT DISCUSSION:

FY2023 Margin Compression:
- Management cited "aggressive pricing strategy" to maintain market share
- Raw material costs (lithium, nickel) increased 15% YoY
- New factory ramp costs in Austin and Berlin

FY2024-2025 Recovery:
- Pricing stabilized; focus shifted to cost reduction
- Manufacturing efficiency improved (vehicles per employee up 12%)
- Energy Storage segment growing faster than Automotive (45% vs 15%)

Risk Factors Evolution:
- FY2023: Added "increased competition from legacy automakers"
- FY2024: Added "regulatory uncertainty on autonomous driving"
- FY2025: Added "supply chain concentration in China"

Plugin Name: discovery-drone

Components:
- Slash Command: /legal:discover
- Sub-agents (specialized by contract type):
    - contract-analyzer-msa.md (Master Service Agreements)
    - contract-analyzer-nda.md (Non-Disclosure Agreements)
    - contract-analyzer-sow.md (Statements of Work)
    - contract-analyzer-saas.md (SaaS/Subscription Agreements)
    - contract-analyzer-license.md (Software Licenses)
- Skills:
    - legal-clause-taxonomy/ (clause identification patterns)
    - risk-scoring/ (severity classification logic)
    - jurisdiction-rules/ (state/country-specific considerations)

/legal:discover indemnification-liability /contracts/vendor-agreements/

{
  "contract": "Acme-Corp-MSA-2023.pdf",
  "clauses": [
    {
      "type": "indemnification",
      "party_indemnifying": "Us",
      "party_indemnified": "Acme Corp",
      "scope": "IP infringement, third-party claims, data breaches",
      "cap": "Unlimited",
      "page": 15,
      "exact_text": "Vendor shall indemnify, defend, and hold harmless Client from any and all claims...",
      "risk_score": "HIGH",
      "risk_reason": "Unlimited indemnification, one-sided (we indemnify them, they don't indemnify us)"
    },
    {
      "type": "liability_limitation",
      "our_cap": "12 months fees",
      "their_cap": "None specified",
      "consequential_damages": "Excluded for us, not for them",
      "page": 16,
      "risk_score": "HIGH",
      "risk_reason": "Asymmetric liability caps favor counterparty"
    }
  ]
}

Plugin Name: voc-engine

Components:
- Slash Command: /product:voc-cluster
- Sub-agents:
    - ticket-themer.md (batch extraction via API)
    - cluster-engine.md (pattern synthesis)
    - impact-scorer.md (prioritization by revenue/churn risk)
- Skills:
    - product-taxonomy/ (feature categories, user personas)
    - sentiment-analysis/ (frustration indicators)
- MCP: Zendesk, Intercom, or Jira (live ticket access)

import anthropic
import json
client = anthropic.Anthropic()
# Load tickets from export
with open('zendesk-export.json') as f:
    tickets = json.load(f)
# Create batch request
batch = client.messages.batches.create(
    requests=[
        {
            "custom_id": f"ticket-{ticket['id']}",
            "params": {
                "model": "claude-sonnet-4-5-20250929",
                "max_tokens": 512,
                "messages": [{
                    "role": "user",
                    "content": f"""Analyze this support ticket.
Ticket #{ticket['id']}:
Subject: {ticket['subject']}
Body: {ticket['description']}
Return JSON only:
{{
  "theme": "one of: mobile, web_ux, integrations, billing,
    performance, onboarding, reporting, api, other",
  "sentiment": "satisfied | neutral | frustrated | angry",
  "pain_point": "1-sentence description of core issue",
  "feature_request": "specific feature if mentioned, else null",
  "churn_risk": "low | medium | high"
}}
"""
                }]
            }
        }
        for ticket in tickets
    ]
)
print(f"Batch ID: {batch.id}")
print(f"Status: {batch.processing_status}")
# Poll for completion, then retrieve results

/product:voc-cluster ticket-themes.json

Plugin Name: voice-dna

Components:
- Slash Command: /marketing:extract-voice
- Sub-agents:
    - tone-analyzer.md (emotional register, formality level)
    - vocabulary-mapper.md (word choices, phrases to use/avoid)
    - pattern-extractor.md (sentence structures, paragraph patterns)
    - example-curator.md (best examples per category)
- Skills:
    - linguistics-frameworks/ (tone dimensions, readability metrics)
    - brand-voice-templates/ (B2B, B2C, technical, casual)

/marketing:extract-voice /content/approved-2024/

Write in [Company]'s brand voice:
- Tone: Confident but approachable, expert but not arrogant
- Avoid: Jargon without explanation, superlatives (revolutionary,
  game-changing), passive voice
- Prefer: Short sentences (15-20 words), concrete examples,
  "you" over "users"
- Structure: Lead with benefit, explain how, end with CTA
- Sample: "Our research shows that teams using automated workflows
  save 10 hours per week. Here's how to set one up in 5 minutes."

Plugin Name: resume-radar

Components:
  - Slash Command: /hr:screen-resumes <job-desc> <resume-folder>
  - Sub-agents:
      - resume-parser.md (parallel, one per resume)
      - skills-matcher.md (compare to requirements)
      - red-flag-detector.md (gaps, inconsistencies)
  - Skills:
      - ats-criteria/ (matching algorithms, scoring weights)
      - bias-mitigation/ (blind screening guidance)
      - skill-taxonomy/ (synonym matching, equivalent skills)

/hr:screen-resumes senior-pm-job-description.pdf /candidates/PM-role-2026/

Extracted Requirements:

REQUIRED SKILLS (Must-Have):
- Product management: 5+ years
- B2B SaaS experience
- SQL proficiency
- Roadmap ownership
- Stakeholder management
- Agile/Scrum certification or experience

REQUIRED EDUCATION:
- Bachelor's degree (any field)
- MBA preferred but not required

NICE-TO-HAVE:
- Fintech industry experience
- Technical background (engineering degree or coding experience)
- Public speaking / thought leadership
- Experience with AI/ML products

LOCATION:
- Remote OK (US timezone overlap required)

{
  "name": "Jane Smith",
  "email": "jane@email.com",
  "location": "San Francisco, CA",
  "linkedin": "linkedin.com/in/janesmith",
  "experience": [
    {
      "title": "Senior Product Manager",
      "company": "Stripe",
      "duration": "2021-2025 (4 years)",
      "achievements": [
        "Launched 3 payment products generating $50M ARR",
        "Led team of 8 engineers, 2 designers",
        "Reduced checkout abandonment 23%"
      ]
    },
    {
      "title": "Product Manager",
      "company": "Plaid",
      "duration": "2018-2021 (3 years)",
      "achievements": [
        "Built bank connection API used by 2000+ apps",
        "Managed $10M product budget"
      ]
    }
  ],
  "skills": ["Product Strategy", "SQL", "Figma", "Jira", "Amplitude", "Python"],
  "education": [
    {"degree": "MBA", "school": "Stanford GSB", "year": 2018},
    {"degree": "BS Computer Science", "school": "MIT", "year": 2014}
  ],
  "certifications": ["Certified Scrum Product Owner (CSPO)"]
}

Match Score = (
  Required Skills Match    x 40%  +
  Experience Threshold     x 30%  +
  Education Fit            x 15%  +
  Nice-to-Have Bonus       x 15%
)

Jane Smith Calculation:
- Required Skills: 6/6 = 100% x 0.40 = 40 points
- Experience: 7 years (exceeds 5+) = 100% x 0.30 = 30 points
- Education: MBA + BS CS = 100% x 0.15 = 15 points
- Nice-to-Have: Fintech (yes), Technical (yes), AI (no) = 67% x 0.15 = 10 points

TOTAL: 95 points → Strong Yes

Plugin Name: board-whisperer

Components:
  - Slash Command: /strategy:board-deck <quarter> <source-folder>
  - Sub-agents:
      - metrics-compiler.md (pull KPIs from multiple sources)
      - narrative-builder.md (synthesize storyline)
      - risk-identifier.md (extract concerns, mitigations)
      - comparison-engine.md (vs. plan, vs. prior period)
  - Skills:
      - board-deck-templates/ (SaaS metrics, growth-stage, enterprise)
      - executive-communication/ (pyramid principle, MECE)
      - financial-modeling/ (ARR calculations, cohort analysis)
  - MCP: Google Slides (direct deck creation)

/strategy:board-deck Q4-2025 /board-prep/Q4-2025/

Q3 Ending ARR:     $38.5M
+ New Business:    +$2.8M
+ Expansion:       +$1.4M
- Churn:           -$0.5M
- Contraction:     -$0.2M
= Q4 Ending ARR:   $42.0M

Net Revenue Retention: 118%
Gross Revenue Retention: 94%
Logo Retention: 92%

Plugin Name: process-surgeon

Components:
  - Slash Command: /ops:audit-process <sop-folder>
  - Sub-agents:
      - process-mapper.md (extract steps, owners, durations)
      - bottleneck-detector.md (identify delays, waste)
      - dependency-analyzer.md (find blocking relationships)
      - benchmark-comparator.md (vs. industry standards)
  - Skills:
      - lean-six-sigma/ (waste categories, improvement frameworks)
      - process-mining/ (cycle time analysis, wait time identification)
      - automation-patterns/ (RPA opportunities, integration points)

/ops:audit-process /sops/customer-onboarding/

CURRENT STATE PROCESS MAP:

Step 1: Sales Handoff (Day 1-2)
  → Sales sends intro email to CS
  → CS reviews deal notes in Salesforce
  → CS schedules kickoff call
  Duration: 2 days | Owner: Sales → CS

Step 2: Kickoff Call (Day 3)
  → Intro meeting with customer
  → Gather requirements
  → Set expectations
  Duration: 1 day | Owner: CS

Step 3: Legal Review (Day 4-10)
  → Customer sends DPA/security questionnaire
  → Legal reviews and responds
  → Back-and-forth negotiations
  Duration: 7 days | Owner: Legal
  BOTTLENECK IDENTIFIED

Step 4: Security Assessment (Day 11-17)
  → Customer IT sends security checklist
  → Our security team responds
  → Remediation if needed
  Duration: 7 days | Owner: Security
  BOTTLENECK IDENTIFIED

Step 5: Technical Setup (Day 18-22)
  → Provision tenant
  → Configure SSO
  → Data migration
  Duration: 5 days | Owner: Implementation

Step 6: Training (Day 23-25)
  → Schedule training sessions
  → Conduct admin training
  → Conduct end-user training
  Duration: 3 days | Owner: CS

Step 7: Go-Live (Day 26)
  → Cutover to production
  → Monitor for issues
  Duration: 1 day | Owner: CS + Support

TOTAL CYCLE TIME: 26 days

Plugin Name: escalation-tamer

Components:
  - Slash Command: /support:deescalate <ticket-id>
  - Sub-agents:
      - thread-summarizer.md (extract key events, timeline)
      - sentiment-tracker.md (emotion trajectory)
      - issue-extractor.md (root cause, customer ask)
      - response-drafter.md (empathetic reply)
  - Skills:
      - crisis-communication/ (de-escalation frameworks)
      - empathy-frameworks/ (acknowledgment patterns)
      - compensation-guidelines/ (when/what to offer)
  - MCP: Zendesk, Intercom, Freshdesk (fetch ticket data)

/support:deescalate TICKET-5678

Plugin Name: dashboard-translator

Components:
  - Slash Command: /data:translate-dashboard <image or data file>
  - Sub-agents:
      - chart-reader.md (extract data from visualizations)
      - trend-detector.md (identify patterns, anomalies)
      - insight-synthesizer.md (so-what, now-what)
      - benchmark-comparator.md (vs. targets, vs. industry)
  - Skills:
      - executive-communication/ (pyramid principle, BLUF)
      - industry-benchmarks/ (SaaS metrics, e-commerce, fintech)
      - statistical-significance/ (when changes matter)

/data:translate-dashboard Q4-metrics-dashboard.png

TRENDS IDENTIFIED:

1. ARR Growth: Accelerating
   - Q1: +8% QoQ
   - Q2: +10% QoQ
   - Q3: +11% QoQ
   - Q4: +12% QoQ
   Pattern: Consistent acceleration over 4 quarters

2. Segment Mix Shift:
   - Enterprise: 25% → 40% of new bookings
   - Mid-market: 50% → 45%
   - SMB: 25% → 15%
   Pattern: Successful upmarket motion

3. Cohort Retention:
   - 2023 cohorts: 85% 12-month retention
   - 2024 cohorts: 91% 12-month retention
   Pattern: Product improvements driving stickiness

4. ANOMALY: Q4 churn spike
   - October: 1.2% monthly churn (normal)
   - November: 2.8% monthly churn (elevated)
   - December: 1.4% monthly churn (normal)
   Investigation needed: November outlier

Use Case: Legal/Finance teams need to know which contracts
are approaching auto-renewal deadlines so they can:
- Renegotiate unfavorable terms
- Cancel unwanted subscriptions
- Budget for renewals

Components Needed:
- Slash command: /contracts:find-renewals <folder> <days-ahead>
- Sub-agent: Contract parser (extract renewal terms)
- Skill: Auto-renewal clause patterns
- Hook: Weekly scheduled scan (optional)
- MCP: Google Calendar (create reminder events)

mkdir -p contract-cliff-detector/.claude-plugin
mkdir -p contract-cliff-detector/commands
mkdir -p contract-cliff-detector/agents
mkdir -p contract-cliff-detector/skills/renewal-patterns
cd contract-cliff-detector

{
  "name": "contract-cliff-detector",
  "version": "1.0.0",
  "description": "Scan contracts for auto-renewal deadlines and alert before it's too late",
  "author": {
    "name": "Your Legal Team",
    "email": "legal@company.com"
  },
  "keywords": ["legal", "contracts", "renewals", "compliance", "procurement"],
  "repository": "https://github.com/yourcompany/contract-cliff-detector"
}

---
description: Scan contracts for upcoming auto-renewal deadlines
arguments: <contract-folder> [days-ahead]
---
# Contract Renewal Scanner
Scan all contracts in the specified folder and identify those with
auto-renewal clauses approaching their notice deadlines.
## Arguments
- **Contract Folder**: Path to folder containing contracts (PDF, DOCX)
- **Days Ahead** (optional): How far ahead to look (default: 90 days)
User provided: $ARGUMENTS
## Process
### Step 1: Inventory Contracts
List all PDF and DOCX files in the specified folder.
Report total count to user.
### Step 2: Parallel Extraction
For each contract, spawn a sub-agent (renewal-extractor) to find:
- Contract parties (us and counterparty)
- Contract type (SaaS, MSA, NDA, etc.)
- Initial term end date
- Auto-renewal clause (yes/no)
- Renewal term length (e.g., 1 year, 3 years)
- Notice period required (e.g., 30 days, 90 days)
- Notice deadline (calculated: term end minus notice period)
### Step 3: Filter and Alert
Filter contracts where:
- Auto-renewal = Yes
- Notice deadline <= today + [days-ahead]
### Step 4: Generate Output
Create Excel workbook with:
**Tab 1: Urgent Renewals** (notice deadline within 30 days)
Columns: Contract, Counterparty, Annual Value, Notice Deadline,
         Days Remaining, Action Required
Conditional formatting: Red if <14 days, Yellow if <30 days
**Tab 2: Upcoming Renewals** (notice deadline within [days-ahead])
Same columns, sorted by deadline
**Tab 3: All Contracts**
Full inventory with renewal terms extracted
**Tab 4: No Renewal Clause**
Contracts without auto-renewal (for reference)
Save as: contract-renewal-alerts.xlsx
### Step 5: Calendar Integration (if MCP available)
For each urgent renewal, create calendar event:
- Title: "CONTRACT RENEWAL DEADLINE: [Counterparty]"
- Date: Notice deadline
- Description: Contract details, action required
- Reminder: 7 days before, 1 day before

---
name: renewal-extractor
description: Extract renewal terms from a single contract
model: sonnet-4.5
tools: [Read]
permissions:
  file_access: read_only
---
You are a contract analyst specializing in renewal terms. Your task
is to extract auto-renewal information from a single contract.
## Instructions
1. Read the entire contract
2. Find the "Term" or "Duration" section
3. Look for auto-renewal language (see patterns below)
4. Extract specific values
5. Calculate notice deadline
## Common Auto-Renewal Patterns
Look for phrases like:
- "automatically renew for successive [period] terms"
- "shall renew unless [party] provides written notice"
- "evergreen clause"
- "continuous renewal"
- "unless terminated with [X] days notice"
## Return Format
Return ONLY valid JSON:
{
  "contract_file": "filename.pdf",
  "parties": {
    "us": "Our Company Name",
    "counterparty": "Vendor/Customer Name"
  },
  "contract_type": "SaaS Agreement | MSA | NDA | SOW | License | Other",
  "effective_date": "2024-01-15",
  "initial_term": {
    "end_date": "2025-01-14",
    "length": "1 year"
  },
  "auto_renewal": {
    "has_auto_renewal": true,
    "renewal_term": "1 year",
    "notice_period_days": 30,
    "notice_deadline": "2024-12-15",
    "exact_clause": "This Agreement shall automatically renew..."
  },
  "financial": {
    "annual_value": 50000,
    "currency": "USD"
  },
  "termination": {
    "can_terminate_for_convenience": true,
    "termination_notice_days": 30
  }
}
If no auto-renewal clause exists:
{
  "contract_file": "filename.pdf",
  "parties": {},
  "auto_renewal": {
    "has_auto_renewal": false,
    "notes": "Contract expires on end date with no auto-renewal"
  }
}

---
name: renewal-patterns
description: Common auto-renewal clause patterns and terminology.
  Use when analyzing contracts for renewal terms.
---
# Auto-Renewal Clause Patterns
## Standard Language Variations
### Pattern 1: Evergreen with Notice
"This Agreement shall automatically renew for successive one (1) year
terms unless either party provides written notice of non-renewal at
least thirty (30) days prior to the end of the then-current term."
Extract:
- Renewal term: 1 year
- Notice period: 30 days
- Notice type: Written
### Pattern 2: Silent Renewal
"Unless terminated in accordance with Section [X], this Agreement
shall continue in effect indefinitely."
Extract:
- This is NOT auto-renewal, it's evergreen
- Look for termination section for notice requirements
### Pattern 3: Price Escalation on Renewal
"Upon renewal, fees shall increase by the greater of (i) 5% or
(ii) the Consumer Price Index change for the prior 12 months."
Extract:
- Renewal terms: Standard
- Flag: Price escalation clause (risk)
### Pattern 4: Multi-Year Commitment
"Initial term of three (3) years, automatically renewing for
additional three (3) year terms."
Extract:
- Renewal term: 3 years (long commitment!)
- Flag: Multi-year auto-renewal (high risk)
## Notice Deadline Calculation
Formula:
Notice Deadline = Term End Date - Notice Period Days
Example:
- Term ends: December 31, 2025
- Notice period: 90 days
- Notice deadline: October 2, 2025
## Risk Flags
Mark as HIGH RISK if:
- Notice period > 60 days (easy to miss)
- Renewal term > 1 year (long commitment)
- Price escalation on renewal
- No termination for convenience
Mark as MEDIUM RISK if:
- Notice period 30-60 days
- Renewal term = 1 year
- Notice must be "written" (vs. email OK)
Mark as LOW RISK if:
- Notice period < 30 days
- Month-to-month after initial term
- Email notice accepted

# Test locally before installing
claude --plugin-dir ./contract-cliff-detector
# In Claude session:
/contract-cliff-detector:find-renewals /contracts/vendor-agreements/ 90

# Install to user scope (personal use)
claude plugin install ./contract-cliff-detector --scope user
# Or install to project scope (share with team via git)
claude plugin install ./contract-cliff-detector --scope project

git init
git add .
git commit -m "Initial contract-cliff-detector plugin"
git remote add origin https://github.com/yourcompany/contract-cliff-detector
git push -u origin main
# Team members install with:
claude plugin install https://github.com/yourcompany/contract-cliff-detector

Coming soon: Anthropic plugin marketplace submission
Follow https://docs.anthropic.com/plugins for updates

User: /sales:research-prospect "Acme Corp"

Claude orchestrates:

1. MCP: Crunchbase
   → Funding: $15M Series A (2024)
   → Investors: Sequoia, a16z
   → Employee count: 120
   → Growth: +50% headcount YoY

2. MCP: LinkedIn Sales Navigator
   → Decision makers: CTO, VP Engineering, VP Product
   → Recent hires: 5 engineers, 2 PMs in last quarter
   → Org chart: Engineering reports to CTO

3. Web scraping (via Browser MCP)
   → Product: B2B SaaS for inventory management
   → Pricing: Not public (request demo)
   → Tech stack: React, Python, AWS (from job postings)

4. MCP: Salesforce
   → Existing record? No
   → Past interactions? No
   → Similar customers? Yes - 3 inventory management companies

User: /finance:categorize-expenses /receipts/Q4-2025/

Claude processes:

Folder contains: 127 receipt images (photos, PDFs, screenshots)

Parallel processing (127 sub-agents):

Each sub-agent:
1. OCR extracts text from receipt image
2. Identifies: Merchant, date, amount, line items
3. Categorizes per company expense policy
4. Flags policy violations

User: /product:synthesize-interviews /research/Q4-interviews/

Claude processes:

Folder contains: 20 interview transcripts (60-90 min each)

Parallel processing (20 sub-agents):

Each sub-agent reads one transcript and extracts:
- Jobs-to-be-done (what customer is trying to accomplish)
- Pain points (frustrations, workarounds)
- Feature requests (explicit asks)
- Quotes (verbatim compelling statements)
- Sentiment (overall satisfaction level)

Synthesis phase:
- Cluster pain points by theme
- Rank by frequency x intensity
- Identify patterns across segments