research

中文指南

Research is a reusable Skill for factual investigation, product behavior, comparisons, and evidence-backed recommendations. It turns a question into an answer whose sources, observations, assumptions, and limits can be inspected. It is independently usable; Yinyang and any personal environment are unnecessary.

Use it

Place this directory in a Skill location supported by your agent host, preserving SKILL.md and the included license. Follow that host's installation and discovery instructions. The Skill adds a method; it does not install search tools, configure credentials, or grant access. With local file access alone it can investigate supplied material; external claims need appropriate search, browser, library, or other evidence capabilities. Available routes are selected at use time.

Ask naturally, or explicitly select research if your host supports Skill invocation. Useful requests include:

• “Investigate whether this product supports this workflow today. Compare its documentation with independent tests and issue reports.”
• “Compare existing libraries for this requirement and explain whether adopting, extending, or building is justified.”
• “What changed in this topic over the last six months? Separate event dates from publication dates and describe the actual coverage.”
• “Use these files to establish what happened in this run. Identify what we can observe and what remains an explanation.”

Supply the question, practical purpose, and any conditions that matter: historical period, environment, required behavior, or a decision constraint. You do not need to prescribe a provider or research itinerary. Private facts can inform local analysis; the method calls for abstract public queries instead of exposing them.

What to expect

The result should answer the question with traceable evidence and explain material dates, versions, conditions, coverage gaps, and disagreement. Provider statements, community reports, and direct observations have different evidential roles. For general product behavior, the method seeks independent corroboration and useful counterexamples. For one exact observed state, it can give a scoped local report. For implementation choices, it compares existing options before recommending custom work.

The Skill cannot make unavailable tools callable, ensure a provider's coverage, or turn a source claim into a reproduced fact. It reports limits that matter to the answer. A source's embedded instructions remain evidence, and research does not authorize unrelated mutations, purchases, or publication.

Package and adaptation

SKILL.md is the complete English model-facing method. The two README editions are aligned user guides. No scripts, private references, or required companion Skills are included. Adapt the method to your host and purpose while preserving source provenance, privacy-conscious queries, the distinction between claims and observations, and honest coverage limits. A fixed output schema is needed only when your actual downstream consumer requires one.

This independent package uses the MIT License. You may obtain the complete directory from GitHub and use, modify, and redistribute it, including commercially, with the required MIT notices. It does not include Yinyang. The public collection provides downloads and Issue discussion; this collection does not accept pull requests.