feat: Context improvements for skills — better context = skills actually work

[elad12390]

Summary

Continuously improve the context quality of skills so they actually work when loaded into AI agents. Better context = skills work = retention. This is core product quality work.

Areas to improve

• Progressive disclosure — skills should expose essential info first, details on demand
• Structured metadata — standardize how skills declare their capabilities, triggers, and constraints
• Context window optimization — ensure skills are formatted to maximize signal-to-noise ratio in limited context windows
• Trigger phrase accuracy — validate that documented trigger phrases actually match how users invoke skills
• Dependency context — when a skill depends on another, provide context about the relationship
• Version-specific context — context should adapt to the installed version

Why this matters

• Product: 5/5, Viral: 2/5
• Core product quality — if skills don't work well, nothing else matters
• This is a continuous effort, not a one-time milestone
• Directly impacts retention: users who have good skill experiences keep using Tank

Priority context

• Phase: Continuous — do continuously, don't roadmap as a single milestone
• "Better context = skills actually work = retention"

Success metrics

• Reduced "skill didn't work" complaints
• Higher skill activation rate (installed → actually used)
• Fewer token wasted on irrelevant skill context