infra-kit.workflow.verify

Standardize post-change verification and “definition of done” evidence for infra work (IaC/Helm/GitHub/Kubernetes).

SKILL.md

GitHub

Verification (Definition Of Done)

Use this skill after implementing changes (or before marking a ticket “done”) to produce a clear, repeatable verification record.

This skill is cross-cutting: it applies regardless of whether the change was Terraform/Terragrunt, Helm, GitHub Actions, or a Kubernetes rollout.

Outcomes

  • Prevent false “fixed” claims by requiring explicit, recorded evidence
  • Make rollback readiness explicit
  • Produce a skimmable verification section suitable for PR description and ticket updates

Where This Fits In The Flow

  • Use after implementation work and before infra-kit.workflow.audit (or as part of the final readiness gate).
  • Use to fill implementation-notes.md and to complete plan.md Verification with real outputs.

Output Rules

  • Always write verification evidence to the repo (not only chat output).
  • Prefer writing into the active work item:
    • docs/infras-kit/work-items/<id>-<slug>/implementation-notes.md
    • and/or add a short verification.md in the same folder when the evidence is large.
  • List only commands that were actually run, and capture the observed outputs (or summaries with links/files).

Workflow

  1. Confirm the “definition of done” inputs:
    • acceptance criteria from ticket.md
    • verification intent from plan.md
  2. Build a verification matrix (AC -> check -> expected -> observed).
  3. Run the strongest deterministic checks available for the change surface:
    • IaC: terraform fmt/validate/plan or terragrunt validate/plan
    • Helm: helm lint, helm template, schema validation (kubeconform), dry-run apply when safe
    • GitHub Actions: actionlint when available; confirm permissions and triggers by inspection
    • Kubernetes runtime: kubectl get/describe/logs evidence (read-only first)
  4. Capture evidence:
    • commands run
    • key outputs (or file paths to captured outputs)
    • what was not run and why (missing creds, missing tools, no cluster access)
  5. Rollout and rollback readiness:
    • what was rolled out (or not)
    • rollback trigger and exact rollback steps
  6. Close with remaining risks and follow-ups.

Verification Record Template

Write the following (at minimum) into implementation-notes.md or verification.md:

  • Scope verified:
  • Acceptance criteria verification matrix:
  • Commands run:
  • Observed evidence:
  • Skipped checks (and why):
  • Rollback readiness:
  • Residual risks:

Guardrails

  • Do not claim “verified” unless you can point to a command output, log excerpt, or explicit inspected file.
  • If verification requires prod access, ask for approval and limit blast radius (read-only first).