feat: add AI defensive-layer cleanup mode and portable harness guidance

[zhouyuanxinand]

Problem

$simplify-codebase already provides a strong evidence-based workflow for ordinary codebase simplification. AI-assisted implementation introduces a second, distinct maintenance burden: implementation-shape tests, build/CI guards, static scans, and inventories that were added only to prevent an AI-generated change from regressing.

Those guardrails need a dedicated investigation and delivery workflow. Treating them as ordinary dead code risks deleting active business, API, security, persistence, lifecycle, or deployment protections. The package also needs a portable SKILL.md contract so the same workflow can be discovered consistently by major coding harnesses.

Proposal

• Keep the existing $simplify-codebase identity and ordinary-simplification workflow intact.
• Add an explicit objective selector for AI defensive-layer cleanup; do not mix its candidates with ordinary simplification in one unreviewed change batch.
• Add focused references for cleanup categories, proof records, stop conditions, and an itemized Handoff deletion ledger.
• Add cross-harness discovery guidance for Codex, Claude Code, Cursor, GitHub Copilot, Cline, Gemini CLI, and OpenCode.
• Add a small portable-contract check for the skill name, frontmatter, and local reference links.

Safety boundaries

Selecting a cleanup category authorizes investigation, not removal of a live contract. Security, credentials, input validation, persistence, lifecycle, real deployment behavior, public APIs, and stored-format compatibility remain protected unless explicitly authorized and independently proven safe to retire.

Acceptance criteria

• Existing $simplify-codebase invocation remains valid.
• The new objective is explicit and routes to category-specific references only when selected.
• A defensive-cleanup Handoff explains each deleted file or material section, its original role, evidence, retained behavior, reintroduction trigger, and verification.
• The package documents portable installation/discovery without vendor-only core frontmatter.
• Structural validation passes.

A follow-up PR implements this proposal against main.