Claude Code

Adoption and communications kit

Champions, launch announcements, drip campaigns, FAQs, pilot feedback and Passion8 data/cache boundaries.

Effective adoption starts with champions demonstrating reusable real-code tasks, then short messages, prompts, FAQs and a stable support channel reduce friction for everyone else.

This page adapts the official Champion and Communications kits into an organizational checklist, adding the data and cache boundaries needed for custom gateways.

#Roles

Role	Responsible for	Not responsible for
Admin/platform team	Providers, accounts, keys, network, settings, MCP, budgets, logs and support	Writing every engineer's prompts
Champion	Real examples, common usage questions, blockers and next users	Improvising legal/security policy
Team lead	Initial tasks, pilot time and retrospectives	Requiring immediate full autonomy
User	Try a real low-risk task and report blockers	Running risky commands without understanding permissions

#Before launch

Check	Why it matters
Shared support channel exists	Keep reusable answers out of scattered private chats
Installation tested on company network	Proxy, CA and firewall issues can affect everyone
Passion8 endpoint/key instructions ready	Avoid extra /v1 in Claude Code's root
Data, ZDR, transcripts and gateway logs explained	Where code goes is an early question
First real task selected	Real bugs/tests convert better than generic demos
Support staffed for 48 hours	Unanswered first problems stop adoption

#Launch announcement template

Subject: Claude Code is available to engineering

Hello everyone,

Claude Code is now available in our engineering environment. This coding agent runs in terminal, IDE, desktop or web, reads repositories, explains code, plans edits and executes commands. It supports workflows beyond autocomplete or chat.

First try:

1. Install following the internal guide.
2. Set the Passion8 root to https://passion8.cc without /v1.
3. Run claude in a real repository.
4. Run /init to create or update CLAUDE.md.
5. Give it a real low-risk task, such as:
   - Investigate a flaky test.
   - Explain an unfamiliar module's call chain.
   - Review the riskiest parts of the current diff.

Security and data boundaries:

- Files, transcripts, MCP, telemetry, debug and gateway logs have separate policies.
- Do not put keys, customer data, production dumps or unapproved services in prompts.
- Follow internal permissions, MCP admission and auto-mode rules.

Post questions in #claude-code. [Owner] is covering support this week.

#Drip campaign

After launch, send one or two short messages weekly to activate specific workflows rather than repeating product introductions.

Timing	Topic	Message
Week 1	/init and CLAUDE.md	Save test commands, directory boundaries and protected files instead of re-explaining the repository.
Week 1	Plan mode	Ask for a file-level plan before approving changes to shared code.
Week 2	@file and @directory/	Supply relevant context directly, such as @src/auth/ or failure logs.
Week 2	Hooks	Use a Stop hook to notify you when long tasks finish.
Week 3	Caching	Avoid frequent /clear during continuous repository work to retain cache reuse.
Week 3	MCP	Use approved servers and OAuth/helpers; no keys in configuration.
Week 4	Auto mode	Limit automation to low-risk repetition; retain hard boundaries for production and secrets.
Week 4	Review	Ask for risks and missing tests in the working diff without unrelated rewrites.

#Champion's first 30 days

Week	Action	Success signal
1	Create channel, pin quickstart, share 2–3 real prompts/screenshots	Questions start; someone reuses a prompt
2	Weekly show-and-tell, public answers and a skill/instruction example	Nonchampions share examples
3	Two or three 15-minute pairing sessions; turn repeated questions into FAQs	Users return voluntarily
4	Recruit another champion and summarize evidence/blockers to leads	Support no longer depends on one person

#FAQ guidance

Question	Suggested answer
What first?	A real low-risk test, explanation or diff review
Trusting edits?	Start with planning and normal human review
It misses repository context	Run /init, improve CLAUDE.md and supply files/directories
It is wrong	Provide failing tests, errors and logs rather than only restating the question
Is data used for training?	Use actual provider/agreement/ZDR/gateway policy, not improvised answers
Can I enable auto mode?	For low-risk tasks after verifying managed policy and tool boundaries

#Pilot feedback

Dimension	Question
Installation	Was the blocker Node, PATH, proxy, CA, login or key?
First task	Was it real and did it produce verifiable output?
Cost	Can users see usage, cache reads/writes and models?
Security	Any sensitive files, MCP, internet, production or approval issues?
Documentation	Which missing answer belongs in internal docs?
Reuse	Can the prompt, skill or instruction snippet serve the team?

#Passion8 boundaries

Topic	Guidance
Claude Code root	https://passion8.cc without /v1
Token	Passion8/enterprise-issued; never committed
Caching	Actual provider/upstream TTL support; inspect usage
ZDR	Does not automatically delete local, MCP, telemetry or gateway logs
Web/mobile	Do not inherit local Passion8 profiles automatically
Support	Collect status, endpoint, model, error code, proxy state and cache fields