AI Enablement Engagement

Session 4 — Context Management & Custom Skills (Detailed Agenda)

Total Duration: 2 hours (120 minutes) Phase / Week: Phase 2, Week 3 or 4 Format: Hands-on workshop, co-authoring in real Turner repos Required attendees: Turner developers; at least one architect; the pilot product's domain expert (committed in Session 0)

Note. If the pilot's repo doesn't exist yet, create a fresh one (even with just a README) so we have somewhere to commit the CLAUDE.md/AGENTS.md/SKILL.md drafts. The skill exercise anchors on a forward-looking workflow such as 'scaffold a new module in our convention' rather than an existing operational runbook.

Pre-session prep (Turner)

  • Identify the workflow on the pilot product that will anchor the SKILL.md exercise (a code-review checklist, a deployment runbook, a domain-specific test pattern, etc.) — surfaced during Sessions 1-2
  • Confirm write access to the pilot product's repo (or a branch/fork) for the live authoring exercise
  • Have the pilot product's domain expert for the chosen workflow on the call

Agenda

0:00 – 0:10 (10 min) — Recap of the governance framework and where context lives

  • Lead: Nubitz
  • Format: Briefing
  • Tie back to Session 2's governance decisions. Reinforce why context assets matter and how they enforce Turner's standards on AI-generated code.

0:10 – 0:30 (20 min) — Anatomy of a good CLAUDE.md / AGENTS.md

  • Lead: Nubitz
  • Format: Demo with annotated examples
  • Walk through 2-3 real examples (sanitized). What goes in, what doesn't, how to keep it maintainable. Address common antipatterns (kitchen-sink files, stale instructions, conflicting signals).

0:30 – 1:00 (30 min) — Exercise A: Author a CLAUDE.md for the pilot product's repo

  • Lead: Developers, architect coordinating; Nubitz floating
  • Format: Hands-on, paired
  • Working in pairs, draft a CLAUDE.md/AGENTS.md for the pilot product. Test it with a small task and observe whether Claude's behavior actually shifts. Iterate.

1:00 – 1:10 (10 min) — Break

1:10 – 1:25 (15 min) — SKILL.md fundamentals

  • Lead: Nubitz
  • Format: Demo
  • What a skill is, when to write one, how it's discovered, and what differentiates a useful skill from one that fights you. Show one example skill end-to-end.

1:25 – 1:50 (25 min) — Exercise B: Build a SKILL.md together

  • Lead: Joint — Nubitz drives, domain expert provides content, developers contribute
  • Format: Co-authoring
  • Build a skill for the chosen pilot product workflow. Validate against 2-3 real tasks on the pilot product. Note adjustments and capture them.

1:50 – 2:00 (10 min) — Maintenance and rollout discussion

  • Lead: Nubitz, facilitating
  • Format: Discussion + synthesis
  • Who owns these files in Turner's org? How do they get reviewed and updated? Capture the operational plan and preview Session 5.

Materials

  • 2-3 sanitized CLAUDE.md example files (Nubitz)
  • A vetted SKILL.md template (Nubitz)
  • Maintenance/ownership one-pager template

What Nubitz produces after this session

  • A committed CLAUDE.md/AGENTS.md in the pilot product's repo (or a PR with one)
  • At least one SKILL.md ready to ship into the pilot product's repo
  • Updates to the Developer Reference Guide covering context-asset authoring

Risks to manage

  • Repo access blockers: without write access this exercise becomes hypothetical — confirm 48 hours in advance.
  • Skill scope creep: pick a narrow,