Skip to content

Team rollout scenario (local trial → CI pilot → release gate)

Use this page if: you want a realistic staged scenario for organizational rollout conversations.

Canonical rollout instructions remain in: Adoption and Recommended CI flow.

This is a recommended adoption flow based on SDETKit's current repository workflows and docs.

It is a scenario, not a claim that a specific company already executed it.

Stage 0 — one engineer proves value locally (impact 1)

Commands:

python -m sdetkit doctor
python -m sdetkit gate fast --format json --stable-json --out build/gate-fast.json

Exit criteria:

  • Team can run commands locally.
  • build/gate-fast.json exists and can be shared in a PR comment.

Stage 1 — CI pilot on pull requests (week 1)

Add one CI job using:

python -m sdetkit gate fast

Exit criteria:

  • PRs consistently produce pass/fail results.
  • Engineers use failed_steps to triage first, logs second.

Stage 2 — introduce policy budgets (week 2)

Add strict thresholds incrementally:

python -m sdetkit security enforce --format json --max-error 0 --max-warn 0 --max-info 0 --out build/security-enforce.json

If initial --max-info 0 is too strict, set a temporary baseline and ratchet down over time.

Exit criteria:

  • Security budget failures are explicit and reviewable in JSON.
  • Team agrees on ratchet plan and owners.

Stage 3 — release lane enforcement (week 3+)

Enable release preflight in release branches/tags:

python -m sdetkit gate release --format json --stable-json --out build/release-preflight.json

Exit criteria:

  • Release decisions reference build/release-preflight.json plus supporting gate artifacts.
  • Handoffs include artifact links instead of only terminal snippets.

Operating rhythm after rollout

  • PR lane: keep gate fast always on.
  • Main/release lane: enforce security budgets and gate release.
  • Incident/failure triage: use artifact-first flow from CI artifact walkthrough.

Anti-patterns to avoid

  • Enabling all strict thresholds at once without baseline discussion.
  • Treating first non-green run as tool failure instead of integration backlog.
  • Reviewing logs before reading structured artifacts.