Claude Chinese Thinking

Problem

When Claude answers complex questions, it often presents only the final result and hides the derivation, trial and error, and edge-case checks. For math problems, algorithm tasks, logic puzzles, and system design decisions, engineers usually need to know why a model reached a conclusion, which assumptions were made explicit, and which alternatives were rejected. claude-thinking-zh addresses this explainability gap: it turns internal reasoning into readable Chinese thinking traces, then refines the final answer into a deliverable solution.

How It Works

The skill separates output into two parts: Thinking and Solution. The first behaves like a lab notebook, preserving exploration, self-correction, assumption flags, and verification impulses. The second behaves like a published paper, removing uncertainty and keeping a clear structure, key evidence, and final answer.

Key steps include:
- Restating the problem: extract known facts, goals, ambiguities, and implicit assumptions.
- Pattern matching: classify the problem as math, logic, algorithm, or system design, then activate the corresponding strategy.
- Example-driven reasoning: validate understanding with a minimal representative case and catch edge-case errors.
- Self-refutation: test counterexamples, alternative interpretations, and boundary conditions after forming a conclusion.
- Structured verification: cover high-risk cases such as empty input, single-element cases, maximum values, and failure modes.

It also selects thinking depth based on uncertainty: simple facts produce a direct solution, while high-risk problems use decomposition, scaffolding, and explicit strategy comparison.

Fit and Limits

It is useful for math word problems, algorithm debugging, logic puzzles, system trade-off analysis, and planning under uncertainty. It is not a good fit for simple factual lookups or speed-first single-step commands. When information is missing, the skill should mark assumptions and continue rather than silently filling gaps.