Skip to main content

Onboarding a repo: readiness, baseline, go-live

Onboarding a repository is one closed settings surface, slipstream_onboard, operated by an agent over the MCP server. Reading it always returns a fresh readiness report; writing it changes the go-live switches. Anything that turns production behavior ON requires explicit confirmation — anything that turns it OFF does not, so rollback is never gated.

Legend: readiness is report-only — a red score never blocks the platform; it tells you exactly what to fix before flipping switches.

The readiness report

Every slipstream_onboard read evaluates eight checks against the live repo and returns pass/fail with a one-line fix per check:

CheckWhat it verifiesSeverity
AllowlistedThe repo is enabled on the platformblocker
Recipe.slipstream.json resolves against the validating schemablocker
Required checkThe default branch requires RunsGreen pinned to the RunsGreen GitHub App (app id 4244269)blocker
Private repoThe repo is private (trust boundary)blocker
Zombie checksEvery other required check context actually runs — a required context that never runs wedges PRswarning
Auto-merge allowed"Allow auto-merge" is enabled in repo settingswarning
Merge commit allowedThe regular merge method is enabledwarning
CI baselineA baseline of your previous CI has been frozeninfo

The rollup is red (any blocker fails), amber (any warning fails or is unknown), or green. The dashboard's fleet view shows the same readiness as a green/amber/red dot per repo. Readiness is informational by design: it never gates webhook intake, runs, checks, or deploys, and Agentic Pipeline never writes to your repo's GitHub settings — readiness only reads.

The frozen baseline

The first time a repository is connected, Agentic Pipeline reads up to 100 successful runs of its configured GitHub Actions workflow and freezes the elapsed-time sample — count, average, median, minimum, maximum. That baseline is immutable: later onboarding calls never re-query or replace it. Every "faster than your old CI" number the dashboard shows compares live Agentic Pipeline runs against that fixed snapshot, so the comparison cannot be quietly re-based.

Go-live switches

The write side of slipstream_onboard covers the closed set of onboarding settings: the platform allowlist, CI mode, ship-on-green mode, live/version URLs for deploy health checks, and the deploy recipe override. Production-enabling writes (turning ship-on-green on, enabling auto-merge) require an explicit confirmation step; rollback, off, and clear writes do not.

Auto-merge is one of those opt-in switches. When enabled for a repo, the moment a CI run records success, Agentic Pipeline merges the open PR whose head commit matches that run — skipping drafts and PRs labeled hold — with a regular merge commit. GitHub branch protection remains the actual guard: a PR that is not otherwise mergeable fails cleanly and the reason is recorded. There is no retry loop; a PR that becomes mergeable later is picked up on its next push.

What your repo ships

Opting in is two files and one branch-protection rule: a recipe in .slipstream.json (CI), optionally a deploy recipe in .slipstream-deploy.json (see ship on green), and the RunsGreen required check on your default branch. PRs, reviews, and merge gates stay exactly as they are.