feat: Pillar 6 — Leave a Trail + skill optimization

[sujalmanpara]

What Changed

Added Pillar 6: Leave a Trail — the missing agentic context layer that makes future agents smarter because a previous agent worked here.

New: Pillar 6 (Leave a Trail)

• Intent comments — WHY + BOUNDARY, never WHAT. Quality over quantity.
• Decision log (DECISIONS.md) — when you choose A over B, record it so the next agent doesn't blindly reverse it
• Architecture maintenance — update docs when structure changes
• Quality Test — one question: will my changes make the next agent faster or slower?

Optimizations

• Task sizing (XS/S/M/L) — a typo fix no longer needs the full 8-step ceremony
• Git state awareness — checks uncommitted changes before branching
• Graceful degradation — proof step adapts to available tooling instead of demanding specific tools
• Trimmed from ~600 to ~280 lines — removed inline bash that duplicated the scripts

Why

The original skill was strong on discipline (discovery, scope, verify) but left nothing behind for future agents. Human best practice is minimal comments + self-documenting code. Agent best practice is the opposite — agents need intent, boundaries, and decision context because they start from zero every session.

Core insight: For humans, the code IS the documentation. For agents, the documentation IS the code.

Decisions Made

• Kept all 3 scripts and stack-conventions reference unchanged (they were solid)
• Merged verification evidence into a flexible format instead of demanding specific tools
• Added sizing to prevent over-engineering simple fixes

Verification

• SKILL.md: valid markdown, all sections cross-referenced
• Scripts: unchanged, already tested
• Line count: 600 → 280 (53% reduction with MORE functionality)