guardrails: scenario and replay harness

[terisuke]

Parent epic: #1

Source brief:

• docs/ai-guardrails/issues/006-scenario-and-replay-harness.md
• packages/opencode/test/scenario/guardrails.test.ts

Problem

Guardrails are only credible if config precedence, plugin behavior, and migration compatibility are exercised automatically.

Deliverables

• scenario tests for managed config precedence
• scenario tests for Claude-compatible skill discovery
• scenario tests for plugin shell environment injection and lifecycle hooks
• follow-up plan for replaying release-gate and provider-admission scenarios

Acceptance

• scenario suite runs under packages/opencode
• the tests are stable on local development machines and CI
• future guardrail issues can link to specific scenario coverage

Notes

• Follow the thin-distribution approach from docs/ai-guardrails/adr/001-thin-distribution-over-deep-fork.md
• Preserve the philosophy imported from claude-code-skills epic feat(guardrails): Wave 8 — review fixes + remaining hooks + multi-model delegation #130: mechanism-first guardrails, fast feedback, pointer-based instructions, and runtime verifiability
• Prefer OpenCode-native config/profile/plugin/command/CI surfaces over core patches

Dependencies

• epic: internal AI guardrails thin distribution for Cor-Incorporated #1
• guardrails: Claude asset inventory and import #3
• docs/ai-guardrails/migration/claude-code-skills-inventory.md

[terisuke]

This issue is in the now-required set for the MVP floor tracked in #16. It should prove the MVP floor in runtime scenarios and replay coverage, not expand into unrelated maturity work.