soia-team/soia-open-skills / skills/soia-meta-prompt-clarity
soia-meta-prompt-clarity
Verified起草、诊断并规格化中英文提示词,保留用户意图、语言与安全边界。触发:「写提示词 / write a prompt」「优化 prompt / improve this prompt」「扩展成可验证规格」
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 4.4 KB |
| agents/openai.yaml | config | 376 B |
| references/english-prompt-authoring.md | markdown | 5.8 KB |
| references/mode-c-disambiguation.md | markdown | 4.6 KB |
| references/mode-d-quality-gate.md | markdown | 2.9 KB |
| references/mode-d-specification.md | markdown | 2.7 KB |
| references/prompt-framework-patterns.md | markdown | 5.0 KB |
references/prompt-framework-patterns.md
View on GitHub# Curated Prompt Framework Patterns
Use this reference only after selecting mode A or B. Named frameworks are optional organizing aids, not a required stage and not evidence that a prompt is high quality.
## Contents
- Selection rules
- Curated patterns
- Domain routing
- Combination limits
- Quality checks
## Selection rules
1. First determine the task, target AI, output, risk, and language.
2. Use no named framework when a direct instruction is already sufficient.
3. Select a framework only when its structure changes the expected behavior.
4. Choose at most one primary framework and one auxiliary framework.
5. Explain the choice in one sentence.
6. Keep mode C safety rules and mode D quality gates outside framework selection.
A framework cannot replace mode C authorization checks, mode D requirement coverage, or either mode's acceptance gates.
## Curated patterns
### APE — Action, Purpose, Expectation
Use for quick tasks where purpose helps the model choose emphasis.
```text
Action → Purpose → Expected result
```
Avoid when the task needs detailed evidence, state management, or authorization boundaries.
### RTF — Role, Task, Format
Use when professional perspective and output shape matter more than background detail.
```text
Role → Task → Output format
```
Skip the role if it adds no real expertise or decision rule.
### RACE — Role, Action, Context, Expectation
Use for medium-complexity work where background materially changes the answer.
```text
Role → Action → Context → Expected result
```
### RASCEF — Role, Action, Steps, Context, Examples, Format
Use for complex but bounded generation tasks with a known workflow and sensitive output format.
Do not force all six fields when examples or prescribed steps would reduce useful flexibility.
### BAB — Before, After, Bridge
Use for persuasive writing that moves an audience from a present problem to a desired state through a credible mechanism.
Do not use for neutral analysis or evidence review.
### Bloom's Taxonomy
Use for learning objectives and assessment design across increasing cognitive depth: remember, understand, apply, analyze, evaluate, create.
Do not treat the levels as a mandatory sequence when the learner already has prerequisite knowledge.
### ELI5
Use to explain a difficult concept to a novice with plain language and concrete analogies.
Do not sacrifice factual accuracy or use a literal five-year-old persona unless the user asks for it.
### Socratic method
Use for guided learning, reflection, or diagnosis through sequenced questions.
Specify whether the AI should ask one question at a time; do not use when the user needs a finished answer immediately.
### Pros and Cons
Use for transparent comparison when trade-offs are central.
Add decision criteria and weights when a recommendation is required; a list alone does not decide.
### What If
Use for scenario analysis, resilience planning, and risk exploration.
Define the baseline and the variables that change between scenarios.
### SCAMPER / HMW
Use SCAMPER for systematic variation of an existing idea; use How Might We for reframing a problem into an open design question.
Do not combine both unless the task explicitly has separate reframing and ideation stages.
### Few-shot
Use one to three examples when exact format, tone, classification boundary, or transformation behavior is difficult to state reliably.
Examples are data, not instructions to inherit beyond the demonstrated pattern. Remove private or irrelevant content.
## Domain routing
| Scenario | Primary candidates | Typical auxiliary |
|---|---|---|
| Quick instruction | APE, RTF | None |
| Professional content | RACE | Few-shot |
| Strict-format generation | RASCEF | Few-shot |
| Persuasive marketing | BAB | RACE |
| Teaching and assessment | Bloom, ELI5, Socratic | Few-shot |
| Decision analysis | Pros and Cons, What If | RACE |
| Creative problem solving | HMW, SCAMPER | RACE |
This table is a shortlist, not a mandatory mapping. Prefer a direct custom structure when it is clearer.
## Combination limits
Good combinations have distinct jobs:
- `RACE + Few-shot`: context and expectation plus format alignment.
- `Bloom + Few-shot`: cognitive level plus assessment examples.
- `What If + Pros and Cons`: scenarios followed by trade-off comparison.
Reject combinations that merely rename the same elements or produce duplicate sections.
## Reasoning and evidence
Do not use “Chain of Thought” as an instruction to reveal hidden reasoning. For complex reasoning, request:
- decomposition into checkable subproblems;
- assumptions and evidence;
- calculations or intermediate artifacts;
- concise rationale;
- counterexamples, uncertainty, and residual risk.
## Quality checks
- Removing the framework name would not make the prompt less understandable.
- Every selected component changes model behavior.
- The framework does not add unsupported facts or permissions.
- The prompt remains proportional to task complexity.
- Mode C and D obligations remain intact.
- The final result is a complete prompt, not a framework worksheet.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |