# 2520 AI Agent Composability Analysis 8ca26161 > AI Agent Composability Analysis - Skill: `tools-only/2520-ai-agent-composability-analysis-8ca26161` (Agent Skill, multi-file: 3 files) - Install (CLI): `npx skillmds@latest add tools-only/2520-ai-agent-composability-analysis-8ca26161` - Raw SKILL.md: https://api.skillmd.com/api/skills/tools-only/2520-ai-agent-composability-analysis-8ca26161/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: tools-only (https://skillmd.com/u/tools-only) - Updated: 2026-09-22 - Page: https://skillmd.com/skills/tools-only/2520-ai-agent-composability-analysis-8ca26161 --- # AI Agent Composability Analysis ## Can AI Agents Be Built By Composing Skills? **Analysis Date:** February 6, 2026 **Source:** [The VC Corner - AI Agent Startup Ideas 2025](https://www.thevccorner.com/p/ai-agent-startup-ideas-2025) **Skills Directory Version:** 0.1.0 (773 skills across 25 domains) --- ## Executive Summary **Answer: YES, many AI agents can be built by composing existing skills**, particularly in B2B SaaS, finance, and e-commerce domains. ### Quick Stats - **Coverage Assessment**: - ✅ **Strong Coverage (50%)**: Sales/Marketing, Finance, E-commerce, Productivity (5 categories) - ⚠️ **Partial Coverage (20%)**: HR/Talent, Creative/Design (2 categories) - ❌ **Gaps (30%)**: Healthcare, Education, Real Estate (3 categories) - **Composability Framework**: - 773 atomic skills with standardized interfaces - 40+ tool categories enable cross-skill data flow - Exit states provide automatic workflow routing - 6 orchestrator skills for multi-domain agents - Security controls built into every skill ### Key Finding The Skills Directory is **fundamentally designed for AI agent composition**. Each skill acts as an independent service with clear inputs/outputs and exit states that naturally chain together. The architecture supports both predefined workflow blueprints (for common patterns) and dynamic composition (for custom agent behaviors). --- ## Table of Contents 1. [Productivity & Workflow Agents](#1-productivity--workflow-agents) 2. [Sales & Marketing Agents](#2-sales--marketing-agents) 3. [HR & Talent Agents](#3-hr--talent-agents) 4. [Healthcare Agents](#4-healthcare-agents) 5. [Finance Agents](#5-finance-agents) 6. [E-commerce Agents](#6-e-commerce-agents) 7. [Education Agents](#7-education-agents) 8. [Real Estate & Construction Agents](#8-real-estate--construction-agents) 9. [Logistics Agents](#9-logistics-agents) 10. [Creative & Design Agents](#10-creative--design-agents) 11. [Composition Framework Deep Dive](#composition-framework-deep-dive) 12. [Concrete Agent Examples](#concrete-agent-examples) 13. [Coverage Matrix](#coverage-matrix) 14. [Key Findings](#key-findings) 15. [Recommendations](#recommendations) --- ## 1. Productivity & Workflow Agents **Relevant Domains**: `superpowers` (14 skills), `development` (5 skills), `ai_ops` (19 skills) **Total Skills**: 38 skills ### Skills Available **Superpowers Domain** (Meta-execution patterns): - `dispatching-parallel-agents` - Coordinate multiple agents in parallel - `subagent-driven-development` - Break tasks into sub-agents - `systematic-debugging` - Structured debugging workflows - `test-driven-development` - TDD workflow orchestration - `using-git-worktrees` - Parallel branch development - `receiving-code-review` - Code review response automation - `verification-before-completion` - Quality gates before delivery **AI Ops Domain** (AI-powered automation): - `ai_agent_performance` - Monitor and optimize AI agent execution - `ai_workflow_automator` - Automate multi-step workflows - `ai_document_parser` - Extract structured data from documents - `ai_meeting_intelligence` - Transcribe, summarize, action items - `ai_ticket_classifier` - Automatic support ticket routing - `ai_response_suggester` - Context-aware response generation ### Composition Pattern **Pattern**: Sequential + Parallel Orchestration **Example Agent: Autonomous Project Manager** ``` 1. Task Breakdown (ai_workflow_automator) input: { project_description } exit: task_list_created 2. Parallel Execution: a) Code Implementation (subagent-driven-development) input: { task_list } exit: implementation_complete b) Documentation (ai_document_parser) input: { code_context } exit: docs_generated 3. Quality Assurance (test-driven-development) input: { implementation, tests } exit: tests_passing OR tests_failing 4. Code Review (receiving-code-review) input: { pull_request } exit: review_addressed 5. Verification (verification-before-completion) input: { final_state } exit: deployment_ready OR needs_revision ``` ### Tools Required - `lifecycle.*` (task tracking, milestone recording) - `ai.generate`, `ai.classify` - `github.*` (PR creation, code review) - `messaging.send_alert` ### Coverage Assessment **✅ Strong Coverage** The Skills Directory has comprehensive productivity and workflow orchestration capabilities through the `superpowers` domain, which provides meta-patterns for agent coordination. Combined with AI Ops automation skills, it can build sophisticated project management, code generation, and workflow automation agents. ### Missing Capabilities - Calendar/scheduling integrations (Google Calendar, Outlook) - Meeting scheduling optimization (FindTime-like) - Resource allocation algorithms - Time tracking integrations --- ## 2. Sales & Marketing Agents **Relevant Domains**: `marketing` (73 skills across 11 subdomains), `revops` (25 skills), `plg` (24 skills), `plg_frameworks` (34 skills) **Total Skills**: 156 skills ### Skills Available **Marketing Domain** (11 subdomains): - **Content** (14 skills): copywriting, content_research_writer, social_content_generator, email_sequence - **CRO** (7 skills): ab_test_setup, page_cro, signup_flow_cro, popup_cro, onboarding_cro - **Outreach** (3 skills): outreach_personalizer, paid_ads, twitter_algorithm_optimizer - **SEO** (3 skills): seo_audit, schema_markup, programmatic_seo - **Strategy** (8 skills): campaign_plan_architect, marketing_ideas, pricing_strategy, launch_strategy - **Analytics** (3 skills): analytics_tracking, developer_growth_analysis, plg_score_explainer - **Research** (6 skills): competitive_ads_extractor, lead_research_assistant, social_listening_analyzer **RevOps Domain** (25 skills): - `lead_qualification` - BANT, MEDDIC, CHAMP frameworks - `opportunity_scoring` - AI-powered deal prioritization - `deal_inspection` - Health checks and risk detection - `next_best_action` - Contextual sales guidance - `pipeline_health` - Forecast accuracy and velocity - `sales_content_recommender` - Battle cards and collateral - `competitive_intel` - Win/loss analysis - `multi_thread_tracker` - Stakeholder mapping **PLG Domain** (24 execution-focused skills): - `activation` - First value moment optimization - `friction_detector` - UX friction identification - `onboarding_flow` - Personalized onboarding - `pql_scoring` - Product-Qualified Lead scoring - `viral_loop` - Organic growth mechanics - `trial_extension_evaluator` - Conversion optimization ### Composition Pattern **Pattern**: Multi-Stage Funnel Orchestrator **Example Agent: Full-Funnel Growth Engine** ```yaml orchestrator: marketing/funnel_optimizer stage_1_acquisition: - marketing/channel_optimizer exit: best_channels_identified - marketing/paid_ads exit: campaigns_launched - marketing/outreach_personalizer exit: sequences_active stage_2_activation: - plg/friction_detector exit: blockers_identified - plg/onboarding_flow exit: onboarding_optimized - plg/activation exit: first_value_delivered stage_3_conversion: - marketing/cro/signup_flow_cro exit: signup_optimized - plg/trial_extension_evaluator exit: trial_extended OR convert_attempted - revops/opportunity_scoring exit: high_priority OR medium_priority stage_4_closing: - revops/deal_inspection (if high_priority) exit: healthy OR red_flags_found - revops/next_best_action exit: actions_recommended - revops/sales_content_recommender exit: content_delivered ``` ### Tools Required - `analytics.*` (events, cohorts, funnels, metrics) - `crm.*` (leads, deals, accounts, pipeline) - `messaging.*` (email, SMS, in-app) - `ai.*` (generate, classify, score, forecast) - `stripe.*` (subscriptions, trials, billing) ### Coverage Assessment **✅ Strong Coverage (Best in Class)** With 156 skills across the entire customer journey, the Skills Directory has unparalleled coverage for sales and marketing agents. It covers: - Full-funnel optimization (acquisition → activation → conversion → closing) - Multi-channel marketing (content, SEO, paid, outbound) - PLG motion (self-serve, trial optimization, viral loops) - Sales enablement (lead scoring, deal inspection, forecasting) - Attribution and analytics ### Missing Capabilities **None** - Comprehensive coverage across all major use cases --- ## 3. HR & Talent Agents **Relevant Domains**: `people_ops` (8 skills) **Total Skills**: 8 skills ### Skills Available **People Ops Domain** (8 skills): - `people_candidate_sourcer` - Automated candidate sourcing - `people_comp_benchmarker` - Compensation benchmarking - `people_dei_tracker` - Diversity metrics tracking - `people_offboarding_manager` - Offboarding workflow automation - `people_onboarding_checklist` - New hire onboarding tasks - `people_perf_review_generator` - Performance review templates - `people_pulse_analyzer` - Employee sentiment analysis ### Composition Pattern **Pattern**: Sequential HR Workflow **Example Agent: Recruitment Assistant** ``` 1. Candidate Sourcing (people_candidate_sourcer) input: { job_description, requirements } exit: candidates_identified 2. Compensation Benchmarking (people_comp_benchmarker) input: { role, level, location } exit: comp_range_determined 3. [MISSING: Interview Scheduling] 4. [MISSING: Interview Evaluation] 5. [MISSING: Offer Generation] 6. Onboarding (people_onboarding_checklist) input: { new_hire_id } exit: onboarding_complete ``` ### Tools Required - `ats.*` (Applicant Tracking System - **NOT AVAILABLE**) - `calendar.*` (scheduling - **NOT AVAILABLE**) - `docusign.*` (offer letters - **NOT AVAILABLE**) - `hr.*` (HRIS integrations - **NOT AVAILABLE**) ### Coverage Assessment **⚠️ Partial Coverage** The Skills Directory has basic HR capabilities (sourcing, compensation, onboarding, offboarding) but lacks: - **Critical gaps** in recruiting pipeline (screening, interviews, offers) - **No ATS integrations** (Greenhouse, Lever, Workday) - **No scheduling** tools for interview coordination - **No assessment** or evaluation frameworks ### Missing Capabilities - Automated interview scheduling - Technical assessment evaluation - Background check orchestration - Offer letter generation and e-signature - Onboarding workflow automation (beyond checklists) - Learning & development recommendations - Performance review cycle automation - Exit interview analysis --- ## 4. Healthcare Agents **Relevant Domains**: None **Total Skills**: 0 skills ### Skills Available **No healthcare-specific skills exist in the Skills Directory.** ### Composition Pattern **N/A** - Cannot build healthcare agents with current skills ### Tools Required (Missing) - `ehr.*` (Electronic Health Records - **NOT AVAILABLE**) - `fhir.*` (Healthcare data standards - **NOT AVAILABLE**) - `hipaa.*` (Compliance and audit - **NOT AVAILABLE**) - `lab.*` (Lab results, diagnostics - **NOT AVAILABLE**) - `scheduling.*` (Patient appointments - **NOT AVAILABLE**) - `billing.*` (Medical billing, insurance - **NOT AVAILABLE**) ### Coverage Assessment **❌ Complete Gap** The Skills Directory has **no healthcare capabilities**. Building healthcare agents would require: - Domain-specific knowledge (medical diagnosis, treatment protocols) - Regulatory compliance (HIPAA, FDA, clinical trial regulations) - Integration with healthcare systems (EMRs, lab systems, imaging) - Patient safety controls and audit trails ### Required Additions for Healthcare Agents 1. **Compliance & Regulatory**: - HIPAA compliance monitoring - Audit trail generation - Patient consent management - Clinical trial protocol adherence 2. **Clinical Workflows**: - Patient intake and triage - Symptom assessment and diagnosis assistance - Treatment plan generation - Medication interaction checking - Lab result interpretation 3. **Administrative**: - Appointment scheduling optimization - Insurance verification and pre-authorization - Medical billing and coding - Claims processing automation 4. **Integrations**: - HL7/FHIR interfaces - EMR/EHR systems (Epic, Cerner) - Lab information systems - Radiology PACS systems --- ## 5. Finance Agents **Relevant Domains**: `finops` (13 skills), `monetization` (20 skills) **Total Skills**: 33 skills ### Skills Available **FinOps Domain** (13 skills): - `burn_rate_monitor` - Cash burn and runway tracking - `scenario_planner` - Financial scenario modeling - `arr_waterfall` - ARR movement analysis - `benchmark_comparator` - Industry benchmarking - `expense_allocator` - Cost center allocation - `gross_margin` - Margin analysis by product/segment - `investor_metrics` - Board-ready financial metrics - `magic_number` - Sales efficiency calculation **Monetization Domain** (20 skills): - `cohort_ltv_analyzer` - Customer lifetime value by cohort - `consumption_analyzer` - Usage-based billing optimization - `discount_optimizer` - Pricing strategy optimization - `dunning_automation` - Failed payment recovery - `entitlement_manager` - Feature access control - `mrr_movement_tracker` - MRR/ARR waterfall analysis - `packaging_optimizer` - Plan/tier optimization - `price_experimentation` - A/B testing pricing - `usage_metering` - Consumption tracking ### Composition Pattern **Pattern**: Real-Time CFO Dashboard + Alerts **Example Agent: Financial Intelligence Agent** ```yaml orchestrator: finops/investor_metrics parallel_data_collection: - finops/burn_rate_monitor input: { period: "current_month" } exit: runway_calculated - finops/arr_waterfall input: { period: "current_quarter" } exit: arr_movements_identified - monetization/mrr_movement_tracker input: { period: "current_month" } exit: mrr_breakdown_complete - finops/gross_margin input: { segment: "all" } exit: margins_calculated conditional_alerts: - if runway < 12 months: trigger: finops/scenario_planner exit: scenarios_generated - if burn_rate increasing: trigger: finops/expense_allocator exit: cost_review_initiated - if churn > target: trigger: monetization/dunning_automation exit: recovery_campaigns_launched aggregation: - ai.synthesize input: { all_financial_metrics } output: { executive_summary, action_items, red_flags } ``` ### Tools Required - `accounting.*` (cash flow, P&L, balance sheet) - `banking.*` (account balances, transactions) - `stripe.*` (subscriptions, invoices, payments) - `analytics.calculate` (financial formulas) - `ai.forecast` (predictive modeling) ### Coverage Assessment **✅ Strong Coverage** The Skills Directory has comprehensive financial operations coverage: - **Cash management**: Burn rate, runway, scenario planning - **Revenue analytics**: ARR/MRR waterfalls, cohort LTV - **Pricing & packaging**: Experimentation, optimization, discount strategy - **Billing operations**: Dunning, entitlements, usage metering - **Investor metrics**: Board-ready dashboards and benchmarks ### Missing Capabilities - Tax compliance and reporting - Multi-currency handling - Equity management (cap table, vesting) - Payroll processing - Accounts payable/receivable automation --- ## 6. E-commerce Agents **Relevant Domains**: `monetization` (20 skills), `plg` (24 skills), `marketing/cro` (7 skills) **Total Skills**: 51 skills ### Skills Available **Monetization** (E-commerce relevant): - `consumption_analyzer` - Cart and purchase behavior - `discount_optimizer` - Dynamic pricing and promotions - `price_experimentation` - A/B test pricing strategies - `payment_method_optimizer` - Payment method recommendations - `usage_metering` - Track product usage and engagement **PLG** (Self-serve purchase flows): - `activation` - First purchase conversion - `friction_detector` - Checkout friction identification - `onboarding_flow` - Product discovery and education - `viral_loop` - Referral and sharing mechanics - `trial_optimization` - Free trial to paid conversion **Marketing/CRO**: - `ab_test_setup` - Experiment framework - `page_cro` - Landing page optimization - `signup_flow_cro` - Registration optimization - `paywall_upgrade_cro` - Conversion optimization - `popup_cro` - Exit intent and promotion popups ### Composition Pattern **Pattern**: E-commerce Conversion Funnel **Example Agent: E-commerce Optimization Engine** ```yaml stage_1_discovery: - marketing/page_cro input: { page_type: "product" } exit: page_optimized - plg/onboarding_flow input: { user_segment } exit: product_education_complete stage_2_cart: - plg/friction_detector input: { funnel_stage: "cart" } exit: friction_points_identified - monetization/discount_optimizer input: { cart_value, user_segment } exit: discount_applied OR no_discount stage_3_checkout: - monetization/payment_method_optimizer input: { country, device } exit: optimal_methods_shown - marketing/cro/popup_cro input: { trigger: "exit_intent" } exit: retention_attempted OR purchase_completed stage_4_post_purchase: - plg/viral_loop input: { order_id } exit: referral_prompted - [MISSING: Post-purchase upsell] - [MISSING: Shipping optimization] ``` ### Tools Required - `stripe.*` (payments, subscriptions) - `analytics.*` (cart abandonment, funnel metrics) - `messaging.*` (abandoned cart emails) - `inventory.*` (**NOT AVAILABLE** - inventory management) - `shipping.*` (**NOT AVAILABLE** - fulfillment) - `recommendation.*` (**NOT AVAILABLE** - product recommendations) ### Coverage Assessment **✅ Strong Coverage (with gaps)** The Skills Directory has excellent coverage for **digital/SaaS e-commerce**: - Conversion funnel optimization (discovery → cart → checkout → post-purchase) - Pricing and promotion optimization - Friction detection and removal - Viral growth mechanics **Gaps for physical goods e-commerce**: - Inventory management and forecasting - Shipping and fulfillment optimization - Returns and exchanges processing - Product recommendations and merchandising - Visual search and discovery ### Missing Capabilities - Inventory forecasting and restocking alerts - Multi-warehouse fulfillment routing - Shipping carrier selection and optimization - Product recommendation engines - Search and merchandising optimization - Return/exchange automation --- ## 7. Education Agents **Relevant Domains**: None **Total Skills**: 0 skills ### Skills Available **No education-specific skills exist in the Skills Directory.** ### Composition Pattern **N/A** - Cannot build education agents with current skills ### Tools Required (Missing) - `lms.*` (Learning Management System - **NOT AVAILABLE**) - `assessment.*` (Testing and grading - **NOT AVAILABLE**) - `curriculum.*` (Content and progression - **NOT AVAILABLE**) - `student.*` (Student information systems - **NOT AVAILABLE**) ### Coverage Assessment **❌ Complete Gap** The Skills Directory has **no education capabilities**. Building education agents would require: - Learning science expertise (pedagogy, assessment design) - Content management (curriculum, lessons, assessments) - Student data privacy (FERPA compliance) - Adaptive learning algorithms ### Required Additions for Education Agents 1. **Adaptive Tutoring**: - Knowledge assessment and gap analysis - Personalized learning path generation - Real-time tutoring and hints - Mastery-based progression 2. **Curriculum Design**: - Learning objective mapping - Content sequencing optimization - Prerequisite dependency tracking - Bloom's taxonomy alignment 3. **Assessment**: - Automated grading (multiple choice, short answer) - Rubric-based evaluation - Formative assessment generation - Learning analytics and insights 4. **Administration**: - Class scheduling and roster management - Attendance tracking - Grade calculation and reporting - Parent/guardian communication --- ## 8. Real Estate & Construction Agents **Relevant Domains**: None **Total Skills**: 0 skills ### Skills Available **No real estate or construction skills exist in the Skills Directory.** ### Composition Pattern **N/A** - Cannot build real estate/construction agents with current skills ### Tools Required (Missing) - `mls.*` (Multiple Listing Service - **NOT AVAILABLE**) - `property.*` (Property data and valuation - **NOT AVAILABLE**) - `permit.*` (Building permits and compliance - **NOT AVAILABLE**) - `project.*` (Construction project management - **NOT AVAILABLE**) - `gis.*` (Geographic information systems - **NOT AVAILABLE**) ### Coverage Assessment **❌ Complete Gap** The Skills Directory has **no real estate or construction capabilities**. Building agents in this space would require: - Domain expertise (property valuation, building codes, zoning) - Physical world integrations (IoT sensors, site inspections) - Regulatory compliance (permits, inspections, certifications) - Geospatial data processing ### Required Additions for Real Estate Agents 1. **Property Search & Discovery**: - MLS integration and search - Property valuation models - Neighborhood analysis - Comparative market analysis (CMA) 2. **Transaction Management**: - Offer generation and negotiation - Contract management - Title search and escrow coordination - Closing document automation 3. **Property Management**: - Tenant screening and placement - Lease management - Maintenance request routing - Rent collection automation ### Required Additions for Construction Agents 1. **Project Planning**: - Cost estimation and bidding - Material takeoff and ordering - Subcontractor coordination - Schedule optimization (Critical Path Method) 2. **Compliance**: - Permit application automation - Building code compliance checking - Inspection scheduling and tracking - Safety compliance monitoring 3. **Site Management**: - Progress tracking with computer vision - Safety incident detection - Material delivery coordination - Quality control inspection --- ## 9. Logistics Agents **Relevant Domains**: None **Total Skills**: 0 skills ### Skills Available **No logistics-specific skills exist in the Skills Directory.** ### Composition Pattern **N/A** - Cannot build logistics agents with current skills ### Tools Required (Missing) - `warehouse.*` (Warehouse management - **NOT AVAILABLE**) - `fleet.*` (Fleet management and routing - **NOT AVAILABLE**) - `shipping.*` (Carrier integration - **NOT AVAILABLE**) - `tracking.*` (Shipment tracking - **NOT AVAILABLE**) - `iot.*` (Sensors, telemetry - **NOT AVAILABLE**) ### Coverage Assessment **❌ Complete Gap** The Skills Directory has **no logistics capabilities**. Building logistics agents would require: - Operations research expertise (routing, scheduling, optimization) - IoT integrations (GPS, sensors, telemetry) - Physical world constraints (vehicle capacity, warehouse space) - Real-time data processing ### Required Additions for Logistics Agents 1. **Supply Chain Forecasting**: - Demand prediction - Inventory optimization - Reorder point calculation - Safety stock management 2. **Fleet Optimization**: - Route optimization (VRP - Vehicle Routing Problem) - Load optimization - Driver scheduling - Fuel efficiency optimization 3. **Warehouse Operations**: - Pick path optimization - Slotting optimization - Labor scheduling - Cross-docking automation 4. **Delivery Orchestration**: - Last-mile delivery optimization - Real-time ETAs - Exception handling (delays, returns) - Proof of delivery automation --- ## 10. Creative & Design Agents **Relevant Domains**: `marketing/media` (1 skill), `marketing/brand` (3 skills), `marketing/content` (14 skills) **Total Skills**: 18 skills ### Skills Available **Marketing/Media**: - `slack_gif_creator` - Animated GIF generation for Slack **Marketing/Brand**: - `humanization_engine` - Brand voice humanization - `interface_design` - UI/UX design assistance - `skene_voice_guardian` - Brand voice consistency **Marketing/Content** (Creative): - `copywriting` - Marketing copy generation - `content_research_writer` - Long-form content creation - `social_content_generator` - Social media posts - `email_sequence` - Email campaign copywriting - `changelog_generator` - Product update communications ### Composition Pattern **Pattern**: Creative Workflow Assistance **Example Agent: Content Marketing Assistant** ``` 1. Research & Strategy - marketing/content_research_writer input: { topic, target_audience } exit: research_complete 2. Brand Alignment - marketing/brand/skene_voice_guardian input: { draft_content } exit: voice_validated 3. Multi-format Content Creation (parallel) a) Long-form Content - marketing/copywriting exit: long_form_complete b) Social Media - marketing/social_content_generator exit: social_posts_created c) Email Campaign - marketing/email_sequence exit: email_series_drafted 4. [MISSING: Design/Visual Assets] 5. [MISSING: Layout and Formatting] 6. [MISSING: Multi-channel Publishing] ``` ### Tools Required - `ai.generate` (text generation) - `ai.edit` (content refinement) - `rag.query` (knowledge retrieval) - `figma.*` (**NOT AVAILABLE** - design tools) - `canva.*` (**NOT AVAILABLE** - visual creation) - `video.*` (**NOT AVAILABLE** - video editing) ### Coverage Assessment **⚠️ Partial Coverage** The Skills Directory has solid **text-based creative capabilities**: - Copywriting and content generation - Brand voice consistency - Multi-channel content (web, social, email) **Gaps in visual/design capabilities**: - No image generation or editing - No design tool integrations (Figma, Adobe, Canva) - No video editing or production - No music/audio creation - No 3D modeling or rendering ### Missing Capabilities - AI image generation (DALL-E, Midjourney style) - Design system management - Asset library management - Video editing and production - Motion graphics and animation - Music composition and audio editing - Collaboration and review workflows (Figma comments, version control) --- ## Composition Framework Deep Dive ### How Skills Compose The Skills Directory uses four primary mechanisms for skill composition: #### 1. Exit States as Handoffs Skills declare named exit states that trigger specific downstream skills. This creates semantic workflow routing rather than just sequential execution. **Example**: `customer_success/health_scoring` ```json { "id": "cs_health_scoring", "exitStates": [ "churn_prediction", // → Triggers churn prediction skill "expansion_playbook", // → Triggers expansion opportunity skill "renewal_orchestration", // → Triggers renewal workflow "idle" // → No action needed ] } ``` **Workflow**: ``` health_scoring → { if health_score < 60: exit(churn_prediction) if health_score > 80 AND expansion_signals: exit(expansion_playbook) if renewal_date < 60 days: exit(renewal_orchestration) else: exit(idle) } ``` #### 2. Orchestration Pattern Meta-skills evaluate context and dispatch to appropriate skills based on business logic. They handle conditional routing, parallel execution, and context management. **Example**: `meta/gtm_orchestrator` ```json { "id": "meta_gtm_orchestrator", "orchestrates": [ "plg/self_serve_expansion", "plg/reverse_trial", "revops/lead_routing", "revops/opportunity_scoring", "ecosystem/partner_influenced_revenue", "ecosystem/nearbound_signal" ], "exitStates": [ "plg_activation", "sales_assist", "lead_qualification", "co_sell_trigger", "partner_sourced", "idle" ] } ``` **Orchestration Logic**: ``` 1. Evaluate account context: - Product usage signals → PLG motion - Enterprise indicators → Sales-led motion - Partner overlap → Partner-sourced motion 2. Route to appropriate motion: if self_serve_capable: trigger plg/self_serve_expansion elif sales_qualified: trigger revops/lead_routing → sales_assist elif partner_overlap: trigger ecosystem/co_sell_trigger 3. Handle parallel actions: - Always run analytics tracking - Optionally trigger nurture campaigns ``` #### 3. Tool Standardization 40+ tool categories are reused across 300+ skills, enabling plug-and-play skill composition. Common tools create natural data flow between skills. **Most Common Tools** (showing composition pattern): | Tool Category | Uses | Enables Composition | | ----------------------- | ---- | ------------------------ | | `lifecycle.get_segment` | 109 | Customer context sharing | | `analytics.track_event` | 74 | Behavior tracking | | `analytics.get_metrics` | 71 | Performance data | | `messaging.send_in_app` | 59 | User communication | | `crm.get_account` | 40 | Account data retrieval | | `crm.update_account` | 38 | Account data writes | | `ai.generate` | 33 | Content creation | | `ai.classify` | 28 | Classification tasks | | `stripe.*` | 75+ | Billing operations | **Example Composition**: ``` Skill 1: health_scoring uses: lifecycle.get_segment → outputs: { health_score, segment_id } Skill 2: churn_prediction uses: lifecycle.get_segment (same segment_id) uses: analytics.get_metrics (behavior data) outputs: { churn_risk, recommended_actions } Skill 3: retention_playbook uses: messaging.send_in_app (send actions to user) uses: crm.update_account (record intervention) ``` The shared `segment_id` from Skill 1 flows into Skill 2, and the `recommended_actions` from Skill 2 flow into Skill 3's messaging. #### 4. Workflow Blueprints Pre-built multi-skill chains with conditional logic, error handling, and parallel execution. Think of these as "agent templates." **Example Blueprint**: `customer_churn_prevention.yaml` (conceptual) ```yaml workflow_id: customer_churn_prevention version: 1.0.0 description: Multi-stage churn prevention workflow steps: - id: step_1 skill: customer_success/health_scoring input: { accountId: '{{context.accountId}}' } timeout: 300s retry: max_attempts: 3 backoff: exponential exit_routing: at_risk: step_2 healthy: step_4 error: handle_error - id: step_2 skill: customer_success/churn_prediction input: healthScore: '{{step_1.output.health_score}}' accountMetrics: '{{step_1.output.metrics}}' timeout: 300s exit_routing: high_churn_risk: step_3 medium_churn_risk: step_4 low_churn_risk: idle - id: step_3 skill: customer_success/retention_playbook input: churnRisk: '{{step_2.output.churn_risk}}' accountContext: '{{step_1.output}}' timeout: 300s outputs: - recommendedActions - stakeholdersToPing exit_routing: playbook_executed: step_4 escalation_needed: handle_escalation - id: step_4 skill: customer_success/nps_followup input: accountId: '{{context.accountId}}' sentiment: '{{step_2.output.sentiment}}' timeout: 300s conditional: 'sentiment < 7' exit_routing: followup_sent: complete skip: complete error_handling: - id: handle_error action: rollback notify: ops_team - id: handle_escalation action: alert notify: csm_manager ``` --- ## Security & Governance Every skill includes security controls appropriate to its risk level: ### Risk Levels | Level | Examples | Controls | | ------------ | --------------------------------- | ---------------------------------- | | **Low** | Copywriting, content generation | Basic sandboxing | | **Medium** | Analytics queries, read-only CRM | Rate limiting, audit logging | | **High** | CRM writes, messaging | Preview mode, approval gates | | **Critical** | Payment operations, data deletion | Two-phase commit, rollback windows | ### Security Controls Example From `customer_success/health_scoring`: ```json { "security_controls": { "payment_controls": { "preview_mode": true, "per_transaction_approval": true, "amount_limits": { "max_per_transaction": 50000, "daily_limit": 200000, "additional_approval_threshold": 25000 }, "rollback": { "enabled": true, "window_seconds": 3600 }, "two_phase_commit": true, "audit_logging": "detailed" } } } ``` ### Execution Environment From `revops/lead_qualification`: ```json { "security_controls": { "execution_environment": { "type": "sandboxed_container", "resource_limits": { "cpu_cores": 1, "memory_mb": 512, "timeout_seconds": 30, "disk_mb": 100 }, "security": { "network_access": false, "filesystem": "read_only", "user": "non_root", "allowed_commands": ["node", "python3", "npm"], "blocked_syscalls": ["exec", "fork", "kill"] } } } } ``` --- ## Concrete Agent Examples ### Example 1: Sales Deal-Closing Agent **Use Case**: Autonomous sales agent that qualifies leads, scores opportunities, and provides deal-closing guidance to sales reps. **Agent Architecture**: ``` ┌─────────────────────────────────────────────────┐ │ Sales Deal-Closing Agent │ ├─────────────────────────────────────────────────┤ │ │ │ 1. Lead Qualification │ │ └─ revops/lead_qualification │ │ input: { leadId, framework: "MEDDIC" } │ │ exit: qualified → 2 │ │ exit: nurture → nurture_campaign │ │ exit: disqualified → archive │ │ │ │ 2. Opportunity Scoring (if qualified) │ │ └─ revops/opportunity_scoring │ │ input: { dealId } │ │ exit: high_priority → 3 │ │ exit: medium_priority → schedule_followup│ │ exit: low_priority → nurture_campaign │ │ │ │ 3. Deal Inspection (if high_priority) │ │ └─ revops/deal_inspection │ │ input: { dealId } │ │ exit: red_flags_found → risk_mitigation │ │ exit: healthy → 4 │ │ exit: stalled → 4 │ │ │ │ 4. Next Best Action (always) │ │ └─ revops/next_best_action │ │ input: { dealContext } │ │ outputs: { │ │ recommendedActions[], │ │ contentRecommendations[] │ │ } │ │ exit: content_needed → 5 │ │ exit: actions_clear → notify_rep │ │ │ │ 5. Sales Content Recommender (if content needed)│ │ └─ revops/sales_content_recommender │ │ input: { dealStage, objections } │ │ outputs: { │ │ relevantContent[], │ │ battleCards[] │ │ } │ │ exit: content_delivered → notify_rep │ │ │ └─────────────────────────────────────────────────┘ ``` **Skills Composition**: | Step | Skill | Input | Output | Exit States | | ---- | ---------------------------------- | --------------------------- | ---------------------------------------------- | -------------------------------------------- | | 1 | `revops/lead_qualification` | `{ leadId, framework }` | `{ qualified, score, frameworkResults }` | qualified, nurture, disqualified | | 2 | `revops/opportunity_scoring` | `{ dealId }` | `{ priorityScore, signals, forecastCategory }` | high_priority, medium_priority, low_priority | | 3 | `revops/deal_inspection` | `{ dealId }` | `{ healthScore, redFlags[], risks[] }` | red_flags_found, healthy, stalled | | 4 | `revops/next_best_action` | `{ dealContext }` | `{ recommendedActions[], contentRecs[] }` | content_needed, actions_clear | | 5 | `revops/sales_content_recommender` | `{ dealStage, objections }` | `{ content[], battleCards[] }` | content_delivered | **Data Flow**: ``` leadId (input) ↓ lead_qualification → { qualified: true, score: 87 } ↓ opportunity_scoring → { priority: "high", signals: ["budget_confirmed", "champion_identified"] } ↓ deal_inspection → { health: 85, redFlags: [] } ↓ next_best_action → { actions: ["Schedule executive briefing", "Send ROI calculator"] } ↓ sales_content_recommender → { content: ["executive_briefing_deck.pdf", "roi_template.xlsx"] } ↓ notify_rep (Slack message with guidance) ``` **Tools Used**: - `crm.get_lead` - Retrieve lead details - `crm.get_opportunity` - Retrieve deal/opportunity data - `crm.get_pipeline` - Fetch pipeline context - `analytics.get_engagement_metrics` - Product usage, website behavior - `ai.score_lead` - AI-powered lead scoring - `ai.classify` - Categorize objections, deal stage - `messaging.send_email` - Automated email sequences (optional) - `messaging.send_alert` - Notify sales rep (Slack/email) **Security Controls**: - **Risk Level**: High (CRM write access) - **Approval Required**: Email sending to prospects, deal stage changes - **Rollback Window**: 30 minutes for CRM updates - **Audit Logging**: All actions logged with timestamp and context - **Sandboxing**: Execution in isolated container with resource limits **Performance**: - **Total Execution Time**: 30-60 seconds (sequential execution) - **Optimization**: Steps 1-2 can run in parallel if account data pre-fetched - **Caching**: Lead/opportunity data cached for 5 minutes **Metrics**: - **Lead-to-Opportunity Rate**: Target > 25% (excellent) - **Opportunity Win Rate**: Target > 30% (excellent) - **Deal Velocity**: Average time from qualified lead to close - **Sales Rep Efficiency**: Time saved per deal (target: 2-3 hours) --- ### Example 2: Customer Churn Prevention Agent **Use Case**: Continuously monitors customer health, predicts churn risk, and automatically triggers retention workflows before customers churn. **Agent Architecture**: ``` ┌─────────────────────────────────────────────────┐ │ Customer Churn Prevention Agent │ ├─────────────────────────────────────────────────┤ │ │ │ ┌─────────────────────────────────────┐ │ │ │ Parallel Health Assessment │ │ │ │ ┌────────────────┐ ┌─────────────┐│ │ │ │ │ Health Scoring │ │ Sentiment ││ │ │ │ │ │ │ Analysis ││ │ │ │ └───────┬────────┘ └──────┬──────┘│ │ │ └──────────┼──────────────────┼───────┘ │ │ └──────────┬───────┘ │ │ ↓ │ │ Risk Aggregation │ │ ↓ │ │ ┌──────────────┴──────────────┐ │ │ │ │ │ │ High Risk Low Risk │ │ │ …(truncated)