deep-research_skill

Overview

This skill performs exhaustive, citation-driven investigations and returns structured, verifiable research reports. It is read-only: it inspects files, configs, and tests and documents findings with precise citations and confidence levels. Use it when decisions depend on complete, auditable evidence rather than quick summaries.

How this skill works

The skill scans the repository breadth-first to map files, key terms, and entry points, then traces important code paths depth-first to collect direct evidence. Every claim is tied to file paths and line numbers; findings include confidence ratings, analysis, and implications. Outputs follow a fixed structure: executive summary, scope, findings with citations, coverage report, and open questions.

When to use it

• Critical decisions require provable evidence
• Delivering documentation or reports that must be auditable
• You need exhaustive coverage with no gaps
• Multiple stakeholders will review or rely on the research
• Tests, configs, or code must be cross-validated before changes

Best practices

• Define scope clearly before scanning to avoid wasted effort
• Prefer direct code citations over inference; mark inferences as Low confidence
• Use breadth-first scan to locate relevant areas, then depth-first trace for each area
• Document side effects, external calls, and test cases with exact file#L ranges
• List open questions and coverage gaps so reviewers can prioritize follow-up

Example use cases

• Validate security-sensitive behavior before deployment with line-cited evidence
• Produce a compliance-ready report showing where features are implemented and tested
• Investigate a regression by tracing entry points and dependencies with citations
• Audit third-party integration code and configuration for unexpected side effects
• Create long-lived documentation that others can verify from cited files and tests