rngneeds-unity（rngneeds Unity 测试）

rngneeds-unity

Use this skill for RNGNeeds-specific work. If the request is only about generic Unity randomness and does not involve RNGNeeds, handle it normally.

Work sequence

1. Identify the job: choose a pattern, write code, explain behavior, or debug a setup.
2. Read only the reference files needed for that job.
3. Ground the answer in actual RNGNeeds behavior and API names.
4. Recommend the right mechanic first when the user describes a gameplay goal but not an implementation.
5. If a detail is missing, inspect the package source or docs instead of inventing methods.

Route by request

• Choose the right RNG pattern: read references/decision-guide.md, then references/common-patterns.md, then references/core-concepts.md as needed.
• Write or refactor Unity/C# code: read references/api-surface.md and references/examples.md.
• Answer tutorial, support, or docs-style questions: read references/guide-workflows.md first, then references/examples.md.
• Explain weird probabilities or surprising results: read references/pitfalls.md and references/core-concepts.md.
• Build dynamic odds from game state: read references/core-concepts.md, references/api-surface.md, and references/examples.md.
• Model decks, draw piles, discard piles, hands, graveyards, limited rewards, or finite stock: read references/decision-guide.md, references/common-patterns.md, references/pitfalls.md, and references/examples.md.

Guardrails

• Distinguish BaseProbability from the final Probability used during selection.
• Distinguish disabled items from removed items, and depletable items from infinite ones.
• Distinguish weighted selection from deck-order operations. If the user cares about top, bottom, hand, discard, cut, or shuffle order, prefer deck extensions over PickValue().
• Warn that disabled items still occupy probability space, and if selection lands on one that pick is ignored. Disabled or depleted states can reduce result count unless MaintainPickCountIfDisabled is enabled.
• Warn that repeat prevention changes behavior: Spread is fast but more biased, Repick is lower bias, Shuffle preserves distribution better but does not guarantee zero repeats between separate picks.
• Warn that influenced lists recalculate before selection, and final displayed outcomes may normalize.
• Warn that PickValues() returns a shared internal list reference. Copy it if the caller must keep results after another pick.
• Prefer TryPickValue(...) when failure matters, especially for value types where default(T) can look like a real result.
• Note that if an item's value implements IProbabilityInfluenceProvider, it overrides an external provider assigned to that item.
• Note that deck extensions are order-centric and use simple seeded index/random operations, not the list's weighted selection methods.
• Prefer KeepSeed or a custom seed provider for deterministic behavior instead of ad hoc randomness hacks.
• Reach for custom selection methods only when built-in behavior does not satisfy the requirement.

Output expectations

• Explain *why* a feature fits the gameplay goal, not only *how* to call it.
• Keep code examples short and focused on RNGNeeds.
• Use RNGNeeds terminology consistently so designers and programmers can discuss the same setup.

Reference map

• references/core-concepts.md: terminology and mental model
• references/decision-guide.md: choose the right feature for the job
• references/common-patterns.md: common gameplay setups
• references/guide-workflows.md: tutorial-style workflows distilled from the docs guides
• references/api-surface.md: high-value classes, properties, and methods
• references/pitfalls.md: gotchas and debugging heuristics
• references/examples.md: compact Unity/C# examples
• references/test-prompts.md: prompt pack for manual skill testing