Module 14 · Lesson 01
Establishing Team Conventions for Claude Use
Reading time: 17 minutes Track: Claude Fluency for Teams · Lead/Manager path
The convention problem
When a team adopts Claude without shared conventions, you get a dozen different approaches to the same tasks. One person writes detailed prompts and gets consistently good output; another writes terse prompts and gets frustrated. One person builds a great prompt library entry for code review; nobody else knows it exists. The team as a whole makes progress much more slowly than it could.
Shared conventions compress the learning curve and compound good practices across the team. Here's what to establish and how.
The minimal viable convention set
Don't try to document everything at once. A small set of well-adopted conventions is worth more than a comprehensive guide nobody reads.
Convention 1: Data classification
What data can team members use with Claude, what needs anonymization, and what never goes in. This is the highest-priority convention — get it right before anything else. (Review Module 9, Lesson 4 if needed.)
One-page summary: Tier 1 (never), Tier 2 (anonymize first), Tier 3 (freely usable). Post it where new team members will find it.
Convention 2: Shared CLAUDE.md
For engineering teams: a team-maintained CLAUDE.md in the repository covers conventions, architecture, and standards. Update it when things change. Everyone benefits from the improvements anyone makes.
For non-engineering teams: a Claude.ai Team space with shared Project instructions and reference documents achieves the same effect.
Convention 3: Prompt library
Identify your 5-10 highest-value recurring prompts. Document them using the template from Module 11, Lesson 4. Designate someone to maintain it. Start small — 5 well-documented prompts are enough to prove value.
Convention 4: Review standards
What review is expected before Claude-generated content is used? Establish the minimum:
- Code: same review as code from any developer
- External communications: human read before send
- Analysis: verify key figures against primary sources
- Decisions: don't rely solely on Claude's analysis for significant choices
Writing this down once removes uncertainty and prevents the two failure modes (no review at all, or excessive paralysis about using Claude).
Running a conventions workshop
The most effective way to build conventions: get the team together for 90 minutes.
Agenda:
- 20 min: Share what's working — everyone demos their most effective Claude use
- 20 min: Share what's not working — frustrations and failure modes
- 30 min: Draft conventions 1-4 above based on the discussion
- 20 min: Assign owners for prompt library and CLAUDE.md
The workshop serves two purposes: it produces the conventions, and it builds shared understanding of how the team is using Claude. Both are valuable.
Communicating conventions to new team members
Add Claude conventions to your onboarding materials:
- Data classification policy (critical — new members need this immediately)
- Link to shared prompt library
- Link to CLAUDE.md / team Projects
- Suggested first tasks to try with Claude
A 30-minute "Claude orientation" for new team members — covering the basics and your specific conventions — pays off within their first week.
Iteration: conventions should change
The conventions you write in month one will be different from the right conventions in month six. Build in a quarterly review: what's working, what's outdated, what new patterns have emerged?
Claude capabilities and team workflows both evolve. Your conventions should too.