Session 4 — Context Management & Custom Skills
Phase: 2 — Enablement & Integration Target Week: Week 3 or 4 Duration: 2 hours Format: Hands-on workshop with co-authoring in Turner repos Audience: Turner development team, architects (recommended)
Note. Authoring CLAUDE.md, AGENTS.md, and SKILL.md is forward-design work — we're encoding what the pilot product will do, not describing existing code. These files seed the pilot's repo with AI-aware conventions before significant code is written.
Objective
Apply the governance framework from Session 2 to the pilot product. Together we author CLAUDE.md, AGENTS.md, and SKILL.md files directly in the pilot product's repos so that AI-generated code reliably reflects Turner's architectural and security standards. The pattern we produce is what Turner will replicate across the rest of their portfolio.
Key Topics
- The role of CLAUDE.md and AGENTS.md: project-level context, conventions, and guardrails
- Anatomy of a good context file: structure, what to include, what to omit, how to keep it maintainable
- SKILL.md files: when to create one, how they're discovered, and how they steer behavior
- Authoring CLAUDE.md and AGENTS.md for the pilot product's repo, live
- Building a SKILL.md for a real workflow on the pilot product (for example: a code-review checklist, a deployment runbook, a domain-specific test pattern)
- Versioning, ownership, and review of context assets — keeping them honest as code evolves
- Validating that context is being applied: quick "smoke tests" developers can run
Outcomes
- The pilot product's repo has a working CLAUDE.md and/or AGENTS.md authored during the session
- At least one SKILL.md file built collaboratively for the pilot product and validated against real tasks
- Pattern/template that Turner can replicate across the rest of the portfolio in Phase 3
- Context-asset entries added to the Developer Reference Guide
Inputs Required from Turner
- Write access to the pilot product's repo (or fork/branch ready for the session)
- The pilot product's domain expert available for the workflow chosen to anchor the SKILL.md exercise