Verified Agent Skill record

deep-research · Nexu

An 8-step research methodology that transforms fuzzy topics into high-quality, deliverable research reports through systematic decomposition, tiered source collection, fact extraction, and independent validation.

Tasks

Classify the research question type and time sensitivity, then decompose the topic into 2-4 researchable sub-questions with explicit boundaries
Collect tiered sources by authority level (L1 through L4), extract verifiable fact cards, and build a comparison framework based on the question type
Derive conclusions by explicitly mapping facts through a comparison framework, submit to an independent agent for validation before reporting, and deliver a final report with a one-sentence summary and traceable references

Inputs

Fuzzy research topics triggered by /deep-research or comparison analysis requests
User-defined research boundaries including population, geography, time period, and hierarchy level

Outputs

00_problem_decomposition.md containing decomposed sub-questions and boundaries
01_source_materials.md containing tiered sources with ongoing updates
02_fact_cards.md containing verifiable fact cards distinguishing official statements from inferences
03_comparison_framework.md containing analysis dimensions selected by question type
04_derivation_process.md containing explicit fact-comparison-conclusion chains
05.5_validation_record.md containing independent agent validation results
05_verification_record.md containing sanity check results from a typical use case
FINAL_research_report.md and ~/outcome.tar.gz containing the final packaged deliverable

Limitations and checks

If a user needs visual diagrams rather than a report, the research-to-diagram skill should be used instead
Community sources (L4) have low credibility and must not serve as the sole support for core conclusions
Unverifiable information must be labeled as source-restricted and cannot be used as the sole basis for any conclusion
Confirm all core conclusions are supported by L1/L2 level facts and that no conclusion has silently upgraded the certainty level stated in the fact cards
Verify that every source in the source materials file maps directly to a decomposed sub-question, removing or demoting any uncorrelated materials
Verify that the independent agent validation in Step 6.5 was executed before final report writing, checking data accuracy, derivation logic, and completeness of comparison dimensions
Confirm that factual descriptions involving enforcement actions strictly use the confirmed tier, and that theoretical risks or official reasons are clearly labeled and not mixed into factual statements
Ensure each chapter of the final report traces back to a sub-question from Step 1, preventing scope creep and knowledge-display digression