Operate CloudForge with clarity and confidence

Write a useful technical prompt

A precise prompt reduces guesswork. You do not need a perfect diagnosis; you need a clear account of what you can observe.

Environment

Subscription scope, region, service tier, runtime, and network model.

Symptom

Exact error, status code, affected operation, and when it began.

Impact

Who or what is affected, and whether production traffic is involved.

Recent change

Deployment, configuration, identity, DNS, certificate, or policy changes.

Evidence

Relevant logs, metrics, correlation IDs, timestamps, and commands already tried.

Environment: Azure App Service, Linux, West Europe, P1v3
Symptom: HTTP 502 began at 14:20 UTC after deployment
Impact: All production requests are failing
Recent change: New startup command and application settings
Evidence: Deployment succeeded; container exits after 30 seconds
Goal: Identify the safest investigation sequence before rollback

Manage investigations

Use one conversation for one problem so evidence, assumptions, and decisions remain connected. CloudForge stores conversation history so you can reopen an earlier investigation and continue from its existing context.

• Use New Chat when you move to a different incident or environment.
• Continue the same chat when new logs or test results belong to the current issue.
• Stop generation when enough information is available or the investigation direction needs correction.
• Do not paste secrets, access tokens, connection strings, or customer-sensitive data.

Ask CloudForge to separate the system into failure boundaries before recommending changes. A good response should tell you what evidence to collect, what each result means, and what to try next.

1. 1

   Confirm scope and timeline

   Establish what is failing, what still works, when it changed, and whether the issue is reproducible.

2. 2

   Separate failure boundaries

   Distinguish application, identity, DNS, network, platform, deployment, and downstream dependency failures.

3. 3

   Collect decisive evidence

   Prefer logs, metrics, health state, configuration snapshots, and timestamped test results over assumptions.

4. 4

   Change one variable

   Use the smallest reversible action that tests the leading hypothesis, then observe the result.

5. 5

   Verify recovery

   Confirm service health, business behavior, monitoring, and any temporary controls after the fix.

Generated commands are a draft for review, not an authorization to change an environment. Ask for an explanation of every parameter and adapt identifiers, scope, and safeguards to your organization.

Supported engineering areas

CloudForge is focused on Microsoft Azure and the operational tools commonly used around it.