kandev/docs/plans/repository-sets/task-10-final-verification.md

id, title, status, wave, depends_on, plan, spec

id	title	status	wave	depends_on	plan	spec
10-final-verification	Final verification and spec reconciliation	done	9	09-public-documentation	plan.md	../../specs/workspaces/requirements/repository-sets.md

Task 10: Final Verification And Spec Reconciliation

Acceptance

• Formatting, backend tests and lint, frontend typecheck, tests, lint, and the i18n gates all pass, and every command's result is recorded in the task file.
• The spec matches the shipped behavior. Any divergence found while building is reflected in docs/specs/workspaces/requirements/repository-sets.md, whose status moves from draft to building, with the same status in docs/specs/INDEX.md.
• plan.md records every task as done, and no task file is left in_progress.
• The change is committed with a Conventional Commits message (feat: ...), staged with explicit paths, after confirming git diff --cached --name-only | grep '\.go$' | xargs gofmt -l is empty.

Verification

cd apps/backend && make fmt && go test ./internal/task/... ./internal/backendapp/... ./internal/events/... ./internal/gateway/websocket/... && make lint
cd apps/web && pnpm run typecheck && pnpm run i18n:check && pnpm run i18n:ratchet
cd apps && pnpm --filter @kandev/web lint
cd apps/web && pnpm vitest run components/task-create-dialog-repository-sets.test.ts lib/state/slices/workspace hooks/domains/workspace app/settings/workspace

Files likely touched

• docs/specs/workspaces/requirements/repository-sets.md, docs/specs/INDEX.md
• docs/plans/repository-sets/plan.md and the task files
• CLAUDE.md or a scoped AGENTS.md, only if a convention documented there changed

Dependencies

All prior tasks.

Inputs

• Root CLAUDE.md: commit conventions, the 100-character commitlint header cap, and the pre-commit prettier/gofmt hooks that fail the commit when they reformat staged files (re-stage and make a new commit rather than amending).
• Do not run /simplify, /qa, /code-review, security review, or a broad /verify here; the PR AI reviewers are the semantic-review gate.

Risks

• The i18n key checker fails on a missing key, an extra key, a dropped placeholder, or a value left identical to English in any of the five locales, so it is the most likely late failure.
• A pre-commit hook reformatting staged files fails the commit; expect one re-stage cycle.

Output contract

Summary, files changed, every command and its result, blockers, risks, divergence from the plan, and final spec/plan/task status updates.