Skip to content

Exercise 06: Design Rules & Skills for a Tiered System

Objective

Design a complete .cursor/ configuration (rules, skills, agents, AGENTS.md) for the FinSecure system from Exercise 05. This tests your ability to architect the configuration layer that governs a tiered multi-agent pipeline.

Required Reading

  • Expert README -- Tiered routing, agent inventory
  • Rules | Cursor Docs -- .mdc format, frontmatter, precedence
  • Agent Skills | Cursor Docs -- Official docs on SKILL.md format, discovery, and activation
  • Agent Skills Guide -- Additional guide on SKILL.md format and activation
  • Review the existing tier routing rule: .cursor-expert/rules/jg-tier-routing.mdc
  • Review the existing skill: .cursor/skills/jg-pipeline-artifact-io/SKILL.md

The configuration architecture you design here applies to Cursor deployments. Rules live in .cursor/rules/*.mdc, skills in .cursor/skills/*/SKILL.md, and agent definitions in .cursor/agents/*.md.

The configuration architecture you design here applies equally to Claude Code deployments. Rules translate to CLAUDE.md directives, skills use the identical .claude/skills/ format, and agent definitions map to model selection in sequential prompting. Design for portability -- the concepts are IDE-agnostic.

Context

In Exercise 05 you designed an architecture proposal for FinSecure, a fintech platform requiring compliance, security review, and cost monitoring. Now design the .cursor/ configuration that would implement that architecture's governance layer.

Tasks

Part 1: Rules Design

Design 3 project rules (.mdc files) for FinSecure. For each rule, specify: - Filename (e.g. jg-security-review-required.mdc) - Full frontmatter (description, alwaysApply, globs if applicable) - Body outline: When to Apply, Rule Content, Exempt

Suggested rules: 1. Tier routing -- complexity classification and agent assignment (reference jg-tier-routing.mdc) 2. Security review required -- mandate jg-reviewer-high for any file in src/auth/, src/payments/, or src/compliance/ 3. Test coverage threshold -- block commits if coverage drops below 80%

Part 2: Skills Design

Design 2 skills (SKILL.md files) for FinSecure. For each skill, specify: - Directory name and SKILL.md frontmatter (name, description) - Body sections (when to use, how to execute, output format, anti-patterns)

Suggested skills: 1. compliance-checker -- validate that changes comply with financial regulations (SOC2, PCI-DSS) 2. cost-estimator -- estimate token/API costs for a pipeline run before execution

Part 3: Agent Inventory

Create a table of agents for FinSecure, including tiered variants. Show: - Agent name, model, tier (fast/standard/high), readonly flag - At least 10 agents covering: planner, subplanner, worker, tester, reviewer, debugger, git, plus domain-specific agents (e.g. team-security-scanner, team-compliance-auditor)

Part 4: AGENTS.md Registry

Write the AGENTS.md content that maps pipeline order to agents, including: - Pipeline execution order (numbered steps) - Subagent type mapping (which subagent_type dispatches to which agent) - Tier routing table (trivial/standard/complex -> agent variants)

Part 5: Activation Flow

Create a mermaid diagram showing how rules, skills, and agents interact during a pipeline run. Show: - When rules are checked (before dispatch, during execution, before commit) - When skills are loaded (on-demand by description match) - How agents read rules and skills to modify their behavior

Output

Write to docs/expert/tutorials/outputs/06-config-design.md with these section headings:

## Rules Design
(Your 3 rule specifications)

## Skills Design
(Your 2 skill specifications)

## Agent Inventory
(Agent table with tiered variants)

## AGENTS.md Registry
(Pipeline order and subagent mapping)

## Activation Flow
(Mermaid diagram of rule/skill/agent interaction)

Validation

python3 docs/expert/tutorials/verify.py --exercise 06

Checks: file exists, all 5 sections present, rules section specifies frontmatter fields and 3+ rule names, skills section specifies SKILL.md format and 2+ skills, agent inventory has a table, document includes a mermaid diagram, total document >= 200 words.

Reflection
  • How do rules and skills interact? Can a rule reference a skill?
  • What happens if two rules conflict (e.g. one says "always use high tier" and another says "use fast for trivial")?
  • How would you version-control this configuration for a team of 10 developers?
Answer

This is a portfolio exercise requiring 5 sections:

  • Rules Design: 3 rules with full frontmatter specs. Key: tier-routing rule, security-review-required rule (with globs for auth/payments/compliance), test-coverage-threshold rule.
  • Skills Design: 2 skills with SKILL.md format. Key: compliance-checker and cost-estimator with specific activation descriptions.
  • Agent Inventory: 10+ agents with tiered variants and domain-specific agents.
  • AGENTS.md Registry: Pipeline order, subagent type mapping, tier routing table.
  • Activation Flow: Mermaid diagram showing when rules check (before dispatch, before commit) and when skills load (on-demand).

See docs/expert/tutorials/solutions/06-config-design-guide.md in the source repo for a complete exemplar.