prompt-engineering
Use only when writing, revising, testing, or versioning prompts, system instructions, developer instructions, few-shot examples, output schemas, refusal behavior, and prompt regression cases.
Works with
--- name: prompt-engineering description: Use only when writing, revising, testing, or versioning prompts, system instructions, developer instructions, few-shot examples, output schemas, refusal behavior, and prompt regression cases. license: MIT --- # Prompt Engineering ## Workflow 1. Define the task, user input shape, desired output, constraints, and failure cases. 2. Keep instructions short, specific, and testable. 3. Use examples only when they materially improve behavior. 4. Prefer structured output when downstream code depends on format. 5. Version prompts and track changes that affect behavior. ## Rules - Separate policy, role, task, context, examples, and output format. - Avoid contradictory instructions. - Do not bury critical constraints in examples only. ## Verification - Run representative prompt tests before and after changes. - Check ambiguity, malformed input, unsafe input, and edge cases.
More AI & ML skills
writing-shape
mattpocock/skills
Writing, exploit: shape raw material into an article, paragraph by paragraph.
writing-fragments
mattpocock/skills
Writing, explore: mine raw fragments, no structure yet.
full-output-enforcement
leonxlnx/taste-skill
Overrides default LLM truncation behavior. Enforces complete code generation, bans placeholder patterns, and handles token-limit splits cleanly. Apply to any task requiring exhaustive, unabridged output.

