← ClaudeAtlas

prompt-enhancementlisted

Prompt engineering and task-specification skill for improving prompts, agent instructions, system messages, evaluation prompts, coding-agent tasks, multi-step workflows, and LLM output reliability. Use when rewriting vague prompts, adding constraints, reducing ambiguity, improving safety, creating reusable prompt templates, or converting goals into actionable agent instructions.
Garyson26/Trident-SecOps-Skills · ★ 0 · AI & Automation · score 68
Install: claude install-skill Garyson26/Trident-SecOps-Skills
# Prompt Enhancement ## Operating Rules - Preserve the user's intent; do not over-constrain creative or exploratory tasks. - Separate objective, context, inputs, constraints, output format, and verification criteria. - Make implicit assumptions explicit when they affect outcome quality. - Prefer testable instructions over personality-heavy wording. ## Workflow 1. Diagnose the prompt for missing context, ambiguous goals, conflicting constraints, and weak success criteria. 2. Rewrite into a direct task with ordered steps and clear boundaries. 3. Add output format only when structure improves usability. 4. Add examples sparingly; use examples that demonstrate edge cases, not just happy paths. 5. Include a short rationale when the user needs to understand tradeoffs. ## Prompt Template ```text Objective: <what outcome is required> Context: <facts, files, audience, constraints> Inputs: <data the model should use> Process: <important reasoning or execution steps> Output: <format, length, tone, artifacts> Validation: <how to check correctness> Boundaries: <what to avoid or ask before doing> ```