← Back to the README
How to take a private copy, edit the rules that drive every decision, and let the agent suggest tuning.
Once you start editing rules, sample inbox data, or running against your real M365 tenant, you'll want a private copy. Public forks cannot be made private on GitHub. Use this repo as a template instead: click Use this template at the top of GitHub and choose Private, or with GitHub CLI:
OWNER=$(gh api user --jq .login) # or override with your org
REPO=my-inbox-agent # or override with any name
gh repo create "$OWNER/$REPO" \
--template Azure-Samples/m365-inbox-serverless-agent-python \
--private --cloneThis creates an independent repo with no fork relationship, so accidental PRs back to Azure-Samples are not possible.
Files most likely to contain personal/tenant information:
skills/vip-rules.md(and the otherskills/*.mdfiles): your VIPs and triage logicsample-data/inbox/*.json: any real mail you paste in for testinglocal.settings.json,.env: secrets and endpoints (already gitignored)infra/main.parameters.json: subscription/tenant-specific values if you customize
Even in a private repo, never commit real secrets. This sample uses managed identity for Foundry and Entra-authorized connectors for Microsoft 365, so there are no app-managed credentials to leak. For any custom integrations you add, keep that pattern (managed identity, then role assignment) rather than pasting keys.
Getting upstream updates. Sync your private copy from this repo with a single GitHub CLI command, then pull locally:
gh repo sync "$OWNER/$REPO" --source Azure-Samples/m365-inbox-serverless-agent-python
git pullThe Functions Serverless Agents Runtime is in preview, so expect occasional fixes worth pulling in. Your edits to skills/, sample-data/, and infra/main.parameters.json will typically merge cleanly.
Edit skills/vip-rules.md to change who counts as a VIP, what should be skipped, and which topics require Teams escalation. Redeploy after changing production rules:
azd deployThe weekly-rule-suggestions agent reviews recent decisions and suggests small policy changes. Treat those suggestions as human-in-the-loop recommendations: copy only the changes you approve into skills/vip-rules.md, review them, then redeploy.