Native Search Research

Use this skill for web research that should stay lightweight, current, and prompt-driven.

Workflow

1. Choose the best-fit category from references/category-profiles.yaml.
2. Read only that category profile, unless the request clearly spans multiple categories.
3. Build a small first-pass query set from the profile's query_focus and query_patterns.
4. Search the web in focused batches, then open only the most promising sources.
5. Prefer primary sources whenever they can answer the question.
6. Synthesize across sources instead of summarizing one page at a time.
7. Include links, note disagreements, and call out dates when freshness matters.

Category Selection

• Use it for software engineering, cloud, security, infrastructure, developer tools, standards, and technical implementation questions.
• Use ai for models, benchmarks, AI products, research papers, provider capabilities, prompting, and agent tooling.
• Use shopping for products, pricing, specs, alternatives, buying advice, and market comparisons.
• Use general for broad current-events-adjacent research, business topics, education, policy, or mixed-source explainers.
• If the user names a category not in the file, create a temporary custom category by following the custom_template shape in references/category-profiles.yaml, state the assumption, and continue.

Search Rules

• Start narrow, then broaden only if the first pass is weak or repetitive.
• Keep search batches small and intentional.
• When the topic is technical, rely on primary sources only: official docs, standards, repositories, papers, or maintainers.
• When the topic is shopping or recommendations, compare multiple current sources and verify important facts such as price, specs, availability, compatibility, or policy details.
• When freshness matters, prefer the newest trustworthy sources and mention exact dates in the answer.
• If sources conflict, do not average them together. Explain the conflict and which source seems stronger.

Synthesis Rules

• Organize the answer around the user's decision or question, not around the order of sources.
• Compare sources on the category's comparison_axes.
• Separate facts, interpretation, and uncertainty.
• Highlight what multiple sources agree on.
• Flag weak evidence, outdated pages, affiliate-heavy content, or missing primary confirmation.
• Keep the final answer concise unless the user asks for a deep dive.

Output Shape

Use this shape when it fits:

1. Direct answer or recommendation
2. Key evidence across sources
3. Important tradeoffs or disagreements
4. Short conclusion
5. Source links

References

• references/category-profiles.yaml