All Features
New

Connect GitHub, deploy a support agent

Link a repository and HelpShelf reads its default-branch README and root package manifest, then creates a reviewable getting-started draft. Use the CLI for deeper code analysis.

From repo to agent in 4 steps

A bounded scan with a review boundary

1

Connect your repo

Sign in with GitHub and select a repository. One click, no configuration files needed.

2

README and manifest are read

The GitHub flow fetches the repository description, default-branch README, and root package.json. It does not traverse routes, components, models, tests, or other docs.

3

A draft is generated

HelpShelf generates a getting-started article as reviewable draft content. Review and publish it before the agent can use it as a source.

4

Re-scan when needed

Use Re-scan now on the Sources page after the README or package manifest changes. The current connection flow does not install a repository webhook.

What the GitHub scan actually reads

Default-branch README
Root package.json presence
Dependency count snapshot
Repository + branch metadata
Commit SHA at scan time
Repository description fallback

Source changed? Re-scan explicitly

The current connection flow does not install a webhook. After updating the README or package manifest, choose Re-scan now on Sources and review the resulting generated draft.

  • Manual Re-scan now action on Sources
  • Default-branch README + root package.json only
  • Generated article remains reviewable draft content
  • Use the CLI for full codebase and CI workflows
README changed
You update the connected branch
Re-scan now
You start the refresh from Sources
Bounded source read
README + package.json fetched
Draft refreshed
Review before publishing

Explicit refresh

The Sources page shows the last GitHub scan and provides Re-scan now. Nothing claims to update until that scan succeeds.

Bounded GitHub source

GitHub sync reads the default-branch README and root package.json only. For routes, schemas, UI strings, tests, and additional Markdown, use the CLI scanner.

Clear data boundary

The scan stores a repository metadata snapshot plus generated draft and bounded source evidence. It does not fetch or persist the rest of the codebase through this flow.

Frequently asked questions

The GitHub flow fetches the selected branch SHA, repository metadata, the default-branch README, and root package.json. It records a metadata snapshot and generates one reviewable getting-started draft with bounded source evidence. It does not traverse routes, components, models, environment variables, tests, or arbitrary repo docs.

The OAuth request currently asks GitHub for repo and read:user scopes. HelpShelf uses that credential to read repository metadata, branch SHA, the root README, and package.json; the scan does not fetch other source files. Generated draft content and bounded source evidence are stored for review. You can revoke the OAuth authorization in GitHub settings.

It does not auto-sync today. The connection flow does not install a GitHub repository webhook. After a source change, open Sources and choose Re-scan now; updated generated content remains behind the review boundary.

There is no self-service disconnect control in the HelpShelf dashboard today. You can revoke the OAuth authorization in GitHub settings to invalidate access. Previously generated HelpShelf drafts remain until you unpublish or delete them.

The bounded GitHub flow is framework-agnostic because it reads only README and package.json. For deeper codebase analysis, run the HelpShelf CLI; that is the path that scans routes, components, tests, and additional documentation.

Ready to connect your repo?

Connect GitHub for a reviewable README-based draft, or use the CLI for a deeper codebase scan.