Agent skill

prompt-optimization

Analyzes and optimizes prompts using BP-001~008 patterns and 3-step flow (detect, optimize, balance). Use when "optimize this prompt", "review prompt quality", "analyze prompt issues", or creating/reviewing rashomon skill content.

View SKILL.md on GitHub Repository
Stars 9
Forks 0

Install this agent skill to your Project

npx add-skill https://github.com/shinpr/rashomon/tree/main/skills/prompt-optimization

SKILL.md

Prompt Optimization Skill

Core Philosophy

  1. Model-Agnostic: Patterns effective across GPT, Claude, Gemini, etc.
  2. Evidence-Based: Based on peer-reviewed research and industry consensus
  3. Actionable: Each detection provides specific, implementable improvements
  4. Non-Destructive: Suggest improvements while preserving user intent and minimizing constraint creep (see references/execution-quality.yaml over_optimization criteria)

Pattern Detection

P1: Critical (Must Fix)

High confidence research evidence for negative impact.

ID Pattern Research Basis
BP-001 Negative Instructions Attention focuses on forbidden content, increasing violation probability. Inverse scaling confirmed
BP-002 Vague Instructions Primary failure cause. 40% of performance variance
BP-003 Missing Output Format Directly linked to hallucination reduction

P2: High Impact (Should Fix)

Consistent improvement when addressed.

ID Pattern Research Basis
BP-004 Unstructured Prompt "Structure > Length" confirmed
BP-005 Missing Context "More context = higher accuracy" confirmed
BP-006 Complex Task Without Decomposition ICLR 2023: 28% error reduction with decomposition

P3: Enhancement (Could Fix)

Incremental improvements in specific contexts.

ID Pattern Research Basis
BP-007 Biased Examples 40% of few-shot effectiveness depends on exemplar selection
BP-008 No Uncertainty Permission Allowing "I don't know" reduces hallucination

3-Step Optimization Flow

Step 1: Initial Analysis

Input: Target prompt Process: Detect patterns (BP-001 through BP-008) Output: .claude/.rashomon/step1-analysis.md

Contents:

  • Detected issues by severity
  • Location in prompt
  • Original prompt preserved

Step 2: Optimization

Input: Step 1 analysis Process:

  • Classify each improvement as Structural, Context Addition, Expressive, or Variance (see Improvement Classification below). Apply only Structural and Context Addition changes.
  • Consolidate redundant improvements
  • Apply in priority order (P1 > P2 > P3) Output: .claude/.rashomon/step2-optimized.md

Contents:

  • Before/after for each change
  • Rationale
  • Optimized prompt

Step 3: Balance Adjustment

Input: Step 2 output Process:

  • Reference references/execution-quality.yaml
  • Confirm all critical aspects are preserved
  • Confirm constraints are proportionate (prompt length increase ≤50%, no constraints that limit valid solutions unnecessarily — see references/execution-quality.yaml over_optimization) Output: Final optimized prompt. Clean up temporary files (.claude/.rashomon/step1-*.md, step2-*.md) after completion.

Conditional Application

BP-004 (Unstructured)

Apply 4-block pattern IF:

  • Prompt longer than 3 sentences
  • Contains multiple distinct instructions
  • Has implicit section boundaries

Skip when:

  • Single simple instruction
  • Already clearly structured
  • Structure would add unnecessary verbosity

BP-006 (Decomposition)

Decompose IF:

  • 3+ distinct objectives
  • Sequential dependencies
  • Each step can be quality-checked

Key Insight: Goal is EVALUABLE GRANULARITY with QUALITY CHECKPOINTS, not decomposition itself.

Improvement Classification

Classification Definition Interpretation
Structural Prompt structure, clarity, specificity improvements Prompt writing technique
Context Addition Project-specific information added from codebase investigation Information advantage
Expressive Different phrasing, equivalent substance Neutral
Variance Within LLM probabilistic variance Original prompt sufficient

Principle: Distinguish between prompt writing improvements (Structural) and information additions (Context Addition).

Reference: references/execution-quality.yaml for detailed criteria.

References

  • references/patterns.yaml - Detailed pattern definitions
  • references/execution-quality.yaml - Quality evaluation criteria
  • references/skills.md - Skill-specific optimization (BP adaptation, 9 editing principles, progressive disclosure, grading)

Recommended Agent Skills

Expand your agent's capabilities with these related and highly-rated skills.

shinpr/rashomon

recipe-eval-skill

Creates or updates Claude Code skills through interactive dialog, then evaluates effectiveness by parallel execution comparison. Use when creating new skills, updating existing skills, or evaluating skill quality.

9 0
Explore
shinpr/rashomon

worktree-execution

Git worktree management for isolated parallel prompt execution. Use when creating isolated environments for prompt comparison or managing worktree lifecycle. Provides creation, cleanup, and orphan detection scripts.

9 0
Explore
shinpr/rashomon

knowledge-base

Project-specific prompt optimization knowledge management. Use when storing or retrieving learned patterns from comparisons. Provides schema, extraction criteria, capacity management, and retention scoring.

9 0
Explore
shinpr/rashomon

recipe-eval-prompt

Compares original and optimized prompts by parallel execution in git worktrees. Use when evaluating prompt improvement effects or learning prompt engineering through concrete examples.

9 0
Explore
shinpr/claude-code-workflows

implementation-approach

Implementation strategy selection framework. Use when planning implementation strategy, selecting development approach, or defining verification criteria.

258 44
Explore
shinpr/claude-code-workflows

recipe-fullstack-build

Execute decomposed fullstack tasks with layer-aware agent routing

258 44
Explore

Didn't find tool you were looking for?

Be as detailed as possible for better results