A useful agent request identifies the website, the desired result, and whether anything should be published. Begin in the preserved source project for code changes, and use a verified MCP connection for account operations.
Inspect before editing#
Inspect this project's Answer Cloud context and list the organizations and sites I can access. Confirm which site matches the public URL I supplied. Explain its kind, release configuration, and content collections. Make no changes yet.
Use this when opening an unfamiliar project, changing accounts, or working with several sites. The assistant should ask you to select the organization if more than one is possible.
Change the website source#
Update the service page using the approved wording below. Preserve the site's visual system and existing URLs. Build locally and show me the result before publishing.
A source change belongs in Astro pages, components, layouts, or styles. The built HTML is an output. Keep the editable source and build instructions with the project so another session can repeat the build.
When ready, give a publication instruction that names the destination:
Publish this reviewed build to staging for the existing site. Verify the returned URL, the changed page, important links, and a grounded chat answer. Report any unresolved warning.
Staging requires the hosted release workflow. A companion uses production publishing instead; the assistant should inspect the site kind and explain that difference.
Draft or publish content#
Update the installation guide from the current repository facts. Reuse the configured collection, save a draft, and give me its private preview. Wait for review before publishing.
After reviewing the preview:
Publish this approved draft to the existing collection and verify the article URL, collection index, and content retrieval.
Explicit approval to publish is sufficient; a second identical confirmation should not be necessary. Draft and publication are separate lifecycle operations, so be clear about the intended state.
Review traffic and customer questions#
Review the last 30 days of analytics for this site. Separate human pageviews, bot requests, referrals from AI services, and chat activity. Explain the reporting window and collection limits before recommending changes.
For conversations:
Analyze customer questions from last month in paginated batches. Use customer messages, summarize recurring unanswered questions, and suggest source-content improvements without publishing them.
Chat tools omit contact details and certain internal metadata; an analysis of available transcripts is not a complete customer-record export.
Recover a release#
Inspect recent deployments for this site and identify the last known good production sequence. Show the current and proposed destination before rolling back. After I approve the recovery, verify the restored public pages.
See deployment history, CMS entries, and analytics for the underlying procedures.