Skip to content

Using the Claude Project

The Trove team has a shared Claude Project loaded with the content from this wiki. This means every team member has access to an AI assistant that already understands our company, processes, policies, and product — without needing to explain the context every time.

How to Access It

  1. Go to claude.ai and sign in with your Elliephant team account
  2. In the left sidebar, click Projects
  3. Open the Elliephant AI Wiki project
  4. Start a new chat — Claude will have full context from the wiki knowledge base

If you cannot see the project, ask Joshua to add you via the Share menu.

What to Use It For

The Claude Project is most useful for:

  • Looking up internal processes — "How do I submit a bug report?", "What's the after-hours escalation process?"
  • Understanding policies — "What does the Acceptable Use Policy say about public Wi-Fi?", "What are the deployment freeze dates?"
  • Getting up to speed on the product — "How does the gift redemption flow work?", "What does the Worker Service do?"
  • Drafting internal documents — using the wiki as context to write emails, summaries, or task descriptions that align with how Trove communicates
  • Onboarding — new team members can ask questions and get accurate, context-aware answers without needing to read the entire wiki

What Not to Use It For

Claude is a useful assistant, but it has limits. Do not rely on it for:

  • Legal or financial decisions — Claude can summarise what a policy says, but it is not legal or financial advice. Escalate these questions to the appropriate person.
  • Answers where the docs might be outdated — if you suspect the wiki has not been updated recently, verify the answer against the source document or ask a team member
  • Sharing externally — the Claude Project contains internal documentation. Do not paste Claude's responses into external communications, client-facing materials, or public channels
  • Final authority on compliance — use the wiki pages and speak to Joshua for anything compliance-related that requires a definitive answer
  • Sensitive personnel matters — do not use Claude to handle or discuss HR issues, performance, or anything confidential

Requesting a Wiki Update

If you notice something in the wiki that is wrong, outdated, or missing, you can request a change. All changes go through a review process before being published.

How to submit a request

  1. Create a task in ClickUp in the Wiki Update Requests list
  2. Include the following in the task:
  3. Which page needs updating (include the URL or page name)
  4. What is wrong or missing
  5. What the correct information should be (if you know it)
  6. Any supporting context or links
  7. Assign the task to Joshua Curci

For urgent corrections (e.g. something that could cause the team to make a wrong decision), also post a message in the #wiki-updates Slack channel with a link to the ClickUp task so it gets seen quickly.

What happens next

  1. Joshua reviews the request
  2. If approved, the change is made and goes through the standard PR process
  3. Once merged, the Claude Project knowledge base is updated so Claude reflects the new content
  4. The ClickUp task is marked as done

When Claude Gets Something Wrong

Claude is only as accurate as the wiki it has been trained on. If Claude gives you an answer that seems incorrect:

  1. Check the source wiki page directly to verify
  2. If the wiki page is also wrong or outdated, submit a change request via ClickUp (see above)
  3. Post in #wiki-updates if it is something the team should be aware of immediately
  4. Do not rely on the incorrect answer in the meantime — flag it and verify with a team member

Tips for Getting Better Answers

  • Be specific — "What are the deployment freeze dates?" works better than "What are the rules around deploying?"
  • Reference the section you're looking for — "According to the Acceptable Use Policy, can I use public Wi-Fi?"
  • Ask for a source — "Which wiki page does that come from?" helps you verify the answer
  • Start a new chat for each topic — Claude's context window is limited; a fresh chat performs better than a long conversation that has drifted