feat: add /slop-detect skill — find AI-generated code that's technically correct but soulless

[HMAKT99]

It compiles. It passes tests. It's still slop.

AI code has tells: the AbstractFactoryProvider used once, the comment that says // Check if user exists above if (user), the 10-line function that should be 3. It works, but it shouldn't exist.

What /slop-detect does

You:   /slop-detect --diff

Claude: AI SLOP FINDINGS
        HIGH   Over-abstraction    helpers.ts — 3 functions each used once
        HIGH   Hallucinated API    ai.ts — calls .streamText() (doesn't exist)
        MED    Obvious comment     login.ts — "// Hash the password" above hash()
        LOW    Verbose code        user.ts — 10 lines that should be 3
        
        SLOP SCORE: 34/100 (Moderate)
        
        0-20: Clean    40-60: Sloppy    80-100: Pure slop

Extends what gstack already does

/plan-design-review already has AI Slop detection for design — letter grades, aesthetic audit, generic-vs-intentional. /slop-detect is the same idea for code: finds over-abstraction, hallucinated APIs, cargo-culted patterns, and verbose-by-default writing.

Slots into the review workflow:

/review          → is the code correct?
/slop-detect     → is the code human-quality?    ← NEW
/ship            → push it

Only .tmpl committed — bun run gen:skill-docs generates the rest.

Test plan

• .tmpl follows template pipeline — uses {{PREAMBLE}}
• Registered in gen-skill-docs.ts (findTemplates), skill-check.ts (SKILL_FILES + TEMPLATES)
• Registered in test/gen-skill-docs.test.ts (ALL_SKILLS), test/skill-validation.test.ts (preamble arrays)
• bun run gen:skill-docs generates valid SKILL.md
• All existing tests pass with skill added