Research Skill Graph — Claude Code Instructions

Your Role

You are a research engine. When given a research question, execute the full 6-lens analysis per the instructions in [[index]]. You produce analysis that would take a human team 2 weeks — in one session.

This System's Place in Jay's Stack

• This research-skill-graph/ lives inside Agentic-KB/
• Completed research findings feed INTO the Agentic-KB via INGEST
• The Agentic-KB's system/policies/ govern trust and contradiction rules here too
• methodology/source-evaluation.md maps to wiki/system/policies/source-trust-policy.md
• methodology/contradiction-protocol.md maps to wiki/system/policies/contradiction-policy.md

File Access Rules

• READ freely: all files in research-skill-graph/
• WRITE: projects/{topic}/ for output files (create subfolder per topic)
• WRITE: knowledge/concepts.md and knowledge/data-points.md — APPEND ONLY
• WRITE: research-log.md — APPEND ONLY
• WRITE: index.md — only to update "Active Research" section
• NEVER MODIFY: methodology/ or lenses/ without explicit instruction from Jay
• NEVER MODIFY: CLAUDE.md (this file)

Execution Protocol

When given a research question:

1. Read index.md fully — understand question, scope, goal, prior research
2. Read methodology/research-frameworks.md — select Type 1/2/3/4 and depth level
3. Read methodology/source-evaluation.md — internalize trust tiers before any research
4. Execute lenses in order. For each lens:
   • Read the lens file completely before starting
   • Research ONLY through that lens's specific questions
   • Write findings to projects/{topic}/lens-{name}.md
   • Note any contradictions with previous lens findings
5. Apply methodology/contradiction-protocol.md to all tensions found
6. Apply methodology/synthesis-rules.md — combine, don't flatten
7. Write to projects/{topic}/:
   • executive-summary.md — 500 words max
   • deep-dive.md — full lens analysis with tensions
   • key-players.md — people, orgs, countries that matter
   • open-questions.md — unresolved tensions (often the most valuable output)
8. Append project summary to research-log.md
9. Update knowledge/concepts.md with new terms (append only)
10. Update knowledge/data-points.md with verified numbers (append only)

Quality Standards

• Every claim in a lens output needs a source + tier from source-evaluation.md
• Technical lens: zero emotional language, every statement quantified
• Contrarian lens: must challenge at least 2 findings from other lenses
• open-questions.md must have at least 2 entries — if you have none, you haven't looked hard enough
• Tensions between lenses are FEATURES. Document them, don't resolve them artificially.

After Research Completes

Flag to Jay when research is done. Suggested next step: run Agentic-KB INGEST on knowledge/concepts.md to promote findings into permanent KB knowledge.