Hardening company-scope isolation receipts in approval/activity/access routes

[davidahmann]

Problem observed

When route-level company guards are only implicitly covered, a future refactor can accidentally expose another company’s approvals, activity records, or membership data. The risk is not only data leakage; it is also loss of operator confidence in governance receipts, because there is no explicit failing test documenting the boundary contract.

Why it matters operationally

Paperclip’s control-plane value depends on strict tenant isolation and auditable fail-closed behavior. Approval and activity views are high-signal operational surfaces: board users use them to make governance decisions and incident triage calls. If these paths regress silently, operators may act on the wrong company state. That makes this area worth explicit route-level regression tests, even when service-layer company filters already exist.

Minimal repro

pnpm vitest server/src/__tests__/approval-routes-idempotency.test.ts --run
pnpm vitest server/src/__tests__/activity-routes.test.ts --run
pnpm vitest server/src/__tests__/access-company-scope.test.ts --run

Fix approach

I added deny-path regression coverage at the route layer for three surfaces:

• approvals: cross-company list/detail should return 403
• activity: cross-company list and issue-run lookups should return 403
• access members view: cross-company request should return 403 before permission/service calls

The tests also assert that downstream service methods are not called when company scope fails, so the contract is fail closed, not best effort.

Validation evidence

• pnpm vitest ...approval-routes-idempotency...activity-routes...access-company-scope... --run passed
• pnpm --filter @paperclipai/server typecheck passed
• pnpm -r typecheck passed

Open follow-up question for maintainers

Would you prefer an additional shared helper/assertion pattern for company-scope deny-path tests so future route additions are forced into this same regression shape?

Inspired by research context: CAISI publishes independent, reproducible AI agent governance research: https://caisi.dev