feat(vercel_ai): detect privileged tools without approval

[CodingWCal]

Change

Adds VAI-013, which flags a Vercel AI tool that shells out or evaluates dynamic code without a tool-level needsApproval gate.

Scope boundary

AI SDK 7 moved approval to call or agent-level toolApproval. This tool-scoped rule cannot verify that setting, so an SDK 7 finding needs manual review. Call or agent-level correlation is future work.

Validation

• Production rule commits: 2bddcde, 537731a
• YAML matches the engine fixture exactly
• Targeted Go test: go test ./internal/rules
• Full Go test suite: go test ./...
• Rulebook check adds no new errors; known baseline remains OAI-112 and PYD-106

Companion PRs

• Engine: feat(vercel_ai): detect privileged tools without approval trustabl#121
• Rulebook: docs(vercel_ai): explain privileged tool approval rule trustabl-rulebook#82

Team

Justin (@jj-javascript): production rule
Calvin: engine integration, scope correction, and release coordination
Karlee (@kperpignant): consulting partner, threat model, and QA review (credited contributor)