Chapter 20: Agents, Skills & Pipelines

Last Updated: 2026-04-16

20.2 The ITI Agent System

The agent system is a coordinated group of specialist AI agents that help build, test, and improve ITI software. Think of them as a software development team where each agent has a defined specialty.

Agent categories

The AGENTS-INDEX.json (v3.0.1, last updated 2026-04-12) catalogs ~95 agents across four categories:

Category	Count	Location
Personal advisors	13	agents/personal/
Executive advisors	12	agents/executive/
Product and domain agents	84	agents/products/ (includes estate professionals, ITI consulting suite, Patriot University, paused agents)
Estate professional agents	13	agents/products/estate-professionals/ (subset of product agents)

Additionally, 34 dev-process agent definitions exist in agents/dev-process/ covering the full software development lifecycle. A separate AGENTS-DEV-PROCESS.json (v1.1.0) indexes these.

Core development agents (dev-process)

Agent	Role	When to Use
Orchestrator	Routes work to specialist agents; coordinates complex multi-agent tasks	Start here — always
Pattern Agent	Architecture and design guidance; recommends patterns from the shared library	Designing new products or features
API Integration	Claude, Tavily, Pinecone, and third-party API integration	Adding new API integrations
Database Agent	Schema design, migrations, query optimization	Designing or modifying database schemas
Template Agent	Project scaffolding; generates boilerplate from shared templates	Creating new products
Migration Agent	Upgrades, refactors, data migrations	Major version bumps, large refactors
Integration Agent	Cross-system integration; connecting products to n8n/Dify	Wiring products to the AI backend
QA Agent	Automated testing; test case design; quality assurance	Testing and bug investigation
Documentation Agent	READMEs, inline docs, architecture docs, CLAUDE.md files	Keeping documentation current
n8n Workflow Engineer	n8n workflow creation and maintenance	Building or debugging n8n workflows
Dify KB Engineer	Dify knowledge base creation and management	Building or maintaining RAG pipelines
Tauri Integration Engineer	Tauri desktop app development	Desktop app features
FastAPI Engineer	FastAPI service development	Python API services
Security/QA Leads	Security review and QA coordination	Pre-release audits
Design track (8 agents)	UI, UX, interaction, accessibility, product design, design systems, visual brand, conversational UI	All design work

Governance agents

Agent	Role
Vibe Coding Guardian	Audits builds against the 15 vibe-coding pitfalls
Context Keeper	Maintains CLAUDE.md accuracy; enforces session protocol
Scope Owner	Evaluates scope changes; manages parking lot
Claims Ombudsman	Audits documents for false or misleading claims
Claims Evidence Curator	Maintains evidence registries; runs staleness checks

Full agent inventory

The complete inventory of agents is in:

• ITI/operations/agents/AGENTS-INDEX.md and AGENTS-INDEX.json (~95 agents)
• ITI/operations/agents/dev-process/AGENTS-DEV-PROCESS.json (34 dev-process agents)

---

20.3 Skills

A skill is a Markdown file that contains a structured procedure an AI assistant follows when invoked. Skills encode specialized knowledge and workflows that would otherwise require extensive prompting to elicit.

Skill structure

---
name: skill-name
description: Third-person description for the AI to understand when to use this skill.
---

# Skill Name

## Instructions

Step-by-step procedure...

## Reference Material

Tables, code examples, configuration snippets...

How to invoke a skill

In Cursor, invoke a skill by describing the task:

“Apply the infrastructure-operations skill and perform a health check.”

Or more directly:

“Use the n8n-workflow-development skill to create a workflow that…”

The Cursor AI reads the skill file and follows its instructions.

How to invoke a skill in the Personal Assistant app

Skills are invoked automatically in the Personal Assistant desktop app. When your message matches skill keywords, up to 2 relevant SKILL.md files are loaded into the advisor’s prompt context.

Example: asking “Help me plan meals for the week” triggers the meal-planning skill. Asking “Evaluate this business proposal” triggers proposal-evaluation and business-proposal-evaluation.

You can also request a specific skill by name:

“Apply the code-review methodology to this function.”

The Macro Advisor has full catalog awareness (249 skills) and can reference any skill. ~110 advisory-relevant skills are keyword-mapped for automatic injection.

Implementation: SKILL_KEYWORDS in constants.ts, matchSkillKeywords() in systemPromptBuilder.ts, load_skill Tauri command in lib.rs.

Skill deployment locations

Location	Count	Contents
ITI/operations/Skills/Claude-Skills/	224	Canonical source for all ITI-authored skills
~/.cursor/skills/	249	Synced ITI skills + 13 Cursor client-builder skills (anylist, apple-music, calendar, chatbot, email, gmail-calendar, google-places, kanban, podcast, rss, todo-list, todoist, youtube)
~/.codex/skills/	236	Synced copies for Claude Code / Codex
~/.cursor/skills-cursor/	11	Cursor platform skills (canvas, create-hook, create-rule, create-skill, create-subagent, etc.)
ITI/.agents/skills/	~224	Antigravity workspace mirror + 6 context docs

SKILLS-INDEX.json (last updated 2026-04-15) tracks 224 skills with category breakdowns.

Always update the canonical source first, then sync to ~/.cursor/skills/ and ~/.codex/skills/.

---

20.4 Agent Pipeline Patterns

When a task requires multiple agents, they are coordinated in a pipeline. Three patterns are documented in ITI/operations/workflow-documentation.md:

Sequential pipeline

One agent’s output becomes the next agent’s input:

Orchestrator
    → Pattern Agent (design architecture)
    → Template Agent (scaffold code)
    → API Integration Agent (wire up APIs)
    → QA Agent (write tests)
    → Documentation Agent (write docs)

Parallel pipeline

Multiple agents work simultaneously on independent tasks:

Orchestrator
    ├── API Integration Agent (Claude API client)
    ├── Database Agent (schema design)
    └── Template Agent (boilerplate)
         ↓ (all complete)
    → Integration Agent (wire everything together)

Collaborative pipeline

Agents review each other’s work:

Pattern Agent (design)
    → QA Agent (review design)
    → Pattern Agent (revise based on QA feedback)
    → Implementation
    → QA Agent (review implementation)

---

20.5 Multi-Agent Orchestration in n8n

For products requiring AI pipelines that involve multiple steps and tools, n8n is the orchestration layer:

n8n Workflow
├── Webhook trigger (user request)
├── Dify retrieval node (RAG context)
├── AI Agent node — Planner (decompose task)
├── Code node — Parse plan into subtasks
├── Sub-workflow calls (parallel HTTP Request nodes)
│   ├── Subtask 1 → AI Agent node
│   ├── Subtask 2 → AI Agent node
│   └── Subtask 3 → AI Agent node
├── Merge node (combine results)
├── AI Agent node — Synthesizer (final response)
└── Webhook response

---

20.6 Creating a New Agent

1. Check ITI/operations/Agents/AGENTS-INDEX.md — a similar agent may already exist.
2. If none exists, create a new agent definition in ITI/operations/Agents//.
3. Follow the agent template from the Template Agent or skills-cursor/create-subagent/.
4. Update AGENTS-INDEX.md and AGENTS-INDEX.json.
5. Update ITI/operations/agents-and-skills.md (global master reference).

---

20.7 Creating a New Skill

1. Check ITI/operations/Skills/SKILLS-INDEX.md — extend an existing skill before creating a new one.
2. If none exists, create a new skill in ITI/operations/Skills/Claude-Skills//SKILL.md.
3. Follow the skill authoring guide (skills-cursor/create-skill/SKILL.md): name, third-person description, instructions, references.
4. Keep skills under 500 lines. Use progressive disclosure (summary first, details second).
5. Sync to ~/.cursor/skills/ per the instructions in CANONICAL-SOURCE.md.
6. Update SKILLS-INDEX.md and SKILLS-INDEX.json.

---

Previous: Chapter 19 — Prompt Engineering | Next: Chapter 21 — Knowledge Bases