Prompt Engineering Standards

A library prompt is not a personal prompt. Personal prompts can be informal, rely on the author's context, and evolve through trial and error. Library prompts must be self-contained, consistently structured, and produce reliable output when used by someone who did not write them. This requires engineering standards that every library prompt must meet.

The standard has five requirements. First: explicit context declaration. Every prompt states what context documents must be loaded before execution. Second: clear input specification. Every variable input is marked with a placeholder and a description of what goes there. Third: output format specification. The prompt defines exactly what the output looks like — structure, length, tone. Fourth: success criteria. What does "good output" look like? How does the user know if the prompt worked? Fifth: known limitations. Where does this prompt fail? What inputs produce unreliable output?

## Prompt: {{PROMPT_NAME}}
**Category:** {{CATEGORY}}
**Version:** {{VERSION}} | **Author:** {{AUTHOR}}
**Last Tested:** {{DATE}}

### Required Context
- [ ] Organization context document (v2+)
- [ ] {{PROJECT_SPECIFIC_CONTEXT}}

### Inputs
- {{CLIENT_NAME}} — the target company name
- {{RESEARCH_QUESTION}} — the specific question to answer
- {{SOURCE_LIST}} — URLs or documents to analyze

### Prompt
[The actual prompt text with {{PLACEHOLDERS}}]

### Expected Output
[Description of format, length, and structure]

### Success Criteria
- Output answers the research question directly
- All claims cite at least one source
- Length is within 800-1200 words

### Known Limitations
- Does not handle non-English sources
- Requires manual verification of financial data