research
Systematic investigation and root cause analysis. Use when debugging persistent issues, understanding complex systems, or before making architectural decisions.
npx skills add majiayu000/claude-skill-registry --skill research-awannaphasch2016-agent-kernel-mcp-2 --agent claude-code
Same command for any agent — swap --agent for codex, cursor, copilot.
Weekly change comes from our own snapshots, not the repository page — it measures attention, not adoption.
What it does
The skill instructs the agent to perform systematic investigation and root cause analysis when debugging persistent issues or before architectural decisions. It emphasizes research after repeated failed fixes, reading primary sources, reproducing locally, inspecting real data, and documenting decisions. It includes a Quick Research Decision Tree, a Meta-Loop workflow with specific tools, and a set of Core Principles guiding when and how to investigate.
How it works
The skill instructs the agent to:
- Apply a decision framework to determine whether to iterate or switch to research, including checks like production incidents, unfamiliar codebases, and architectural decisions.
- Use a loop pattern with tools such as
/observe,/hypothesis,/research,/validate, and/reflectto form and test hypotheses and synthesize learnings. - Follow Core Research Principles: begin with Read Primary Sources, reproduce locally, and inspect real data; test minimal examples and verify data shapes; avoid relying on blog posts for critical decisions.
- Conduct a structured Research Workflow and adhere to an Investigation Checklist and Boundary Verification when distributed systems are involved.
- Perform Architectural Investigations with an OWL-based relationship analysis framework, defining concepts, applying relation types (Part-Whole, Complement, Substitution, Composition), and documenting concrete examples and trade-offs.
- Apply Common Research Scenarios, including Type System Mismatch, Production Incidents, and Unfamiliar Codebases, with concrete command examples and data verification steps.
- Avoid anti-patterns such as blind iteration without research and reliance on secondary sources.
When to use it
The skill signals usage for cases such as a bug persisting after multiple fixes, needing to understand an unfamiliar codebase, investigating production incidents, making architectural decisions, debugging complex interactions, or learning new technology/library. It explicitly states when not to use it, including the first fix attempt, well-understood problems, or time-critical incidents.
What it can touch
The skill lists tools like AWS CLI, Git, ripgrep, jq, curl, and browser DevTools; specific examples reference commands and data inspection steps (e.g., open docs, test with minimal examples, use curl/jq to inspect responses, and git history commands). It prescribes using Read official docs, Inspect real data, and Test minimal example as part of the workflow.
Caveats
It emphasizes reading primary sources and verifying against ground truth, warns against relying on blog posts or AI without verification, and documents post-incident analysis workflows. It includes explicit cautions about avoiding unnecessary iterations and ensuring steps are grounded in evidence before deploying fixes.
# Research Skill **Tech Stack**: AWS CLI, Git, ripgrep, jq, curl, browser DevTools **Source**: Extracted from CLAUDE.md investigation principles and debugging patterns. --- ## When to Use This Skill Use the research skill when: - ✓ Same bug persists after 2+ fix attempts - ✓ Need to understand unfamiliar codebase - ✓ Investigating production incidents - ✓ Making architectural decisions - ✓ Debugging complex system interactions - ✓ Learning new technology/library **DO NOT use this skill for:** - ✗ First fix attempt (try the obvious solution first) - ✗ Well-understood problems (just fix it) - ✗ Time-critical incidents (fix first, investigate later) --- ## Quick Research Decision Tree ``` What's the problem? ├─ First time seeing this issue? │ ├─ YES → Try obvious fix (iteration) │ └─ NO → Same bug after 2 attempts? → RESEARCH │ ├─ Production incident? │ ├─ Affecting users NOW? → Rollback/hotfix first, research later │ └─ Post-incident analysis? → Deep research │ ├─ Need to understand codebase? │ ├─ Specific function/module? → Read code + tests │ ├─ System architecture? → Trace request flow │ └─ Historical context? → Git blame + commit history │ ├─ Technology decision? │ ├─ Read offici
- When to Use This Skill
- Quick Research Decision Tree
- Loop Pattern: Meta-Loop → Initial-Sensitive
- Core Research Principles
- Principle 1: Research Before Iteration
- Principle 2: Read Primary Sources
- Principle 3: Reproduce Locally
- Principle 4: Inspect Real Data
- Research Workflow
- Investigation Checklist
- Boundary Verification
- Architectural Investigations
- Common Research Scenarios
- Scenario 1: Type System Mismatch
ChatGPT might hallucinate or give outdated answer
open https://pymysql.readthedocs.io/en/latest/
python3 << 'EOF'
Step 1: Extract minimal failing case
Instead of testing entire Lambda function...
aws lambda invoke --function-name worker --payload '{}' /tmp/response.json
Wait 30 seconds for CloudWatch logs...
curl -s https://api.example.com/tickers | jq . > sample_response.json
cat sample_response.json
Inspect structure:What does the research skill do?
Systematic investigation and root cause analysis. Use when debugging persistent issues, understanding complex systems, or before making architectural decisions.
How do I install it?
Run `npx skills add majiayu000/claude-skill-registry --skill research-awannaphasch2016-agent-kernel-mcp-2 --agent claude-code` — it drops the skill into your project so the agent can pick it up. Swap the --agent value for codex, cursor or copilot if you use one of those.
Where does this skill come from?
From majiayu000/claude-skill-registry, a repository with 534 stars. We read it straight from the repository tree rather than a submitted listing, so what you see here is what is actually published.
Is a popular skill a good skill?
Not necessarily. Stars measure attention, not adoption — a repository can trend for a week and be abandoned. That is why we show the weekly change from our own snapshots next to the total, instead of a single flattering number.