Refactor SKILL.md for token efficiency (~48% reduction)

[huntclubhero]

Summary

• Refactor SKILL.md from ~3,500 words to ~1,800 words (~48% token reduction)
• Extract detailed content to reference files following progressive disclosure pattern
• Add two new reference files for on-demand loading

Changes

SKILL.md (lean core)

• Condensed prose into tables where appropriate
• Removed embedded examples (already exist in examples/)
• Kept essential extraction workflow and trigger conditions
• Added clear pointers to reference files

New: resources/research-workflow.md

• Detailed web research strategies
• Search query templates
• Citation format guidance

New: resources/quality-guidelines.md

• Quality gates checklist
• Anti-patterns to avoid
• Skill lifecycle (creation → deprecation)
• Self-reflection prompts
• Memory consolidation guidance

Rationale

This follows Claude Code skill development best practices for progressive disclosure:

1. Metadata (~100 words) - Always in context
2. SKILL.md body (~1,800 words) - Loaded when skill triggers
3. Reference files (unlimited) - Loaded only when Claude needs details

The original SKILL.md loaded everything at once, consuming unnecessary context tokens for detailed reference material that's only needed occasionally.

Token Impact

File	Before	After
SKILL.md	~3,500 words	~1,800 words
Context on trigger	~3,500 words	~1,800 words
Detailed refs	(embedded)	On-demand

Net savings: ~1,700 words per skill invocation (~48%)

Test plan

• Verify skill triggers correctly on /claudeception
• Confirm reference files load when detailed guidance needed
• Test extraction workflow produces valid skills
• Verify existing examples still referenced correctly

---

🤖 Generated with Claude Code