Module Overview

Module 6 is where you stop re-explaining your standards every session and start encoding them. You learn the three universal customization concepts — rules (always-on project guidance), agents (named scoped AI roles), and skills (on-demand domain expertise) — all as portable markdown that transfers across tools. The point is leverage: build the asset once, and every future session and every future cohort benefits.

What You’re Building

By the end of this module, you will have built and contributed three assets to the project’s knowledge directory. These are not exercises — they are real, version-controlled assets that every future session and every future cohort participant benefits from.

Part 1 — Building a Rule

Rules are instructions the AI sees at the start of every session. You write them once; every session benefits. AGENTS.md is the cross-tool baseline — read automatically by most AI coding tools. Tool-specific override files layer on top when you need tool-specific behavior. See the Appendix for wiring details per tool.

What Makes a Good Rule

Build Your Rule — E→P→A→V

Part 2 — Building an Agent

An agent is a prompt that’s been promoted to a saved, named, reusable role. One job per agent. Tight scope, restricted tools, explicit constraints.

The Agent Markdown Format

---
name: agent-name-in-kebab-case
description: One sentence — precise enough to drive auto-delegation.
tools: Read, Grep, Glob    # only what the agent needs
model: claude-opus-4-7     # optional override
---
 
# Agent Name
 
You are [role description]. [What you know. What project you’re
working on. What standards apply.]
 
## What you do
[Specific responsibilities and behaviors]
 
## What you never do
[Explicit constraints — safety, scope, tools]
 
## Output format
[Exact format specification — structure, fields, severity labels]

Design Principles

Build Your Agent — E→P→A→V

Part 3 — Building a Skill

Skills are packaged domain expertise loaded dynamically when a task matches. They keep the main context clean while making specialized knowledge available on demand. The SKILL.md format is an open standard — the portable artifact is a markdown file that works across tools.

The SKILL.md Format

---
name: skill-name-in-kebab-case
description: One sentence — precise enough to drive dynamic loading.
---
 
# Skill Name
 
## What this skill provides
[Domain expertise — standards, conventions, checklists]
 
## Inputs
[What the AI receives when this skill is loaded]
 
## What to do
[Step-by-step or checklist instructions]
 
## Output format
[Exact output specification]
 
## Constraints
[What this skill explicitly does not cover]

Build Your Skill — E→P→A→V

Part 4 — Governance

The more autonomy you give an agent, the more important its constraints are. Governance is about designing for the failure cases, not just the happy path.

Tool Scope as Governance

Autonomy Levels

"Never Do" Rules to Consider

Every well-designed agent has an explicit constraints section. Start with these and add role-specific ones:

## What you never do
 
- Never hard-delete records. Only soft-delete with is_deleted = 1.
- Never expose password_hash, reset_token, or reset_expires in output.
- Never suggest architectural changes — flag for human review.
- Never commit or push code — leave that to the human.
- Never modify files outside [specified directory].
- Never accept requirement changes — document for stakeholder sign-off.
- Never log or echo PII in output, even in error messages.
- Never include real client data in agent prompts stored in the repo.

Role-Specific Build Guide

Same framework, different domain. Use the E→P→A→V prompts from Parts 1–3 — the table below tells you what to build for your role.

Deliverable Quality Bar

Each asset must pass its own Evaluate → Plan → Apply → Validate cycle. Here is what "pass" means for each:

Due

Week 3 Friday — presented and contributed before Module 7. All three assets committed to knowledge/ under your role’s subfolder. Included in the Sprint 1 Readiness gate review alongside code and tests.

Self-Check

• I can explain the difference between rules, agents, and skills — and when to use each.
• I’ve audited AGENTS.md and identified at least 2 rule additions for my role.
• I’ve built a custom agent with tight scope, tool restrictions, system prompt, never-do section, and output format.
• I’ve built a skill with precise description, step-by-step instructions, and output format.
• I’ve tested both assets against real tasks and validated role-quality output.
• I’ve wired my agent and skill into my primary AI tool.
• I understand governance: tool scope, never-do sections, autonomy levels, privacy guardrails.
• All three assets are committed to knowledge/ under my role’s subfolder.

Appendix — Tool-Specific Wiring Reference

The formats (AGENTS.md, agent markdown, SKILL.md) are portable across tools. How you load them varies. This appendix covers wiring for the primary AI tools. The methodology and the assets are the same — only the loading mechanism changes.

Rules

Agents

Skills