guardrails: Claude asset inventory and import

[terisuke]

Parent epic: #1

Source brief:

• docs/ai-guardrails/issues/002-claude-asset-inventory-and-import.md
• docs/ai-guardrails/migration/claude-code-skills-inventory.md

Problem

The Claude harness contains reusable assets, but they are not yet classified by what OpenCode can consume directly versus what must be redesigned.

Deliverables

• migration inventory for skills, rules, hooks, and scripts
• direct-import path for Claude-compatible skills
• AGENTS.md plus instructions layout for long-form rules
• hook mapping table with target implementation type: plugin / command / CI gate / drop

Acceptance

• the inventory names each hook and assigns a migration bucket
• at least one representative .claude/skills/*/SKILL.md fixture is exercised by scenario tests
• repo guidance explains when to keep assets in .claude versus move them into .opencode

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
• docs/ai-guardrails/adr/003-claude-assets-migrate-by-role.md
• docs/ai-guardrails/migration/claude-code-skills-inventory.md