Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Prompt Optimization · aiskillstore | DecimalAI

aiskillstore/prompt-optimization

SkillScore95Maintained

Expert prompt optimization for LLMs and AI systems. Use when building AI features, improving agent performance, crafting system prompts, or optimizing LLM interactions. Masters prompt patterns and techniques.

AI/MLby aiskillstore

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

AI Engineer pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/aiskillstore-prompt-optimization/SKILL.md

decimalai skills pull aiskillstore-prompt-optimization --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/aiskillstore-prompt-optimization/SKILL.md https://app.decimal.ai/s/aiskillstore-prompt-optimization/SKILL.md

Raw file: /s/aiskillstore-prompt-optimization/SKILL.md ↗ · machine-readable summary at /s/aiskillstore-prompt-optimization.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

Runs

—

no usage data yet

Used when offered

—

offered 1,211 · activation not measurable for prompt-injection usage

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

22 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 8d ago

Yes — +9% lift · at higher cost+44% tokens0% turns95% of whole cases pass with skill (21/22)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-04	✗→✓	▲ Improved	137%	0%
case-08	✗→✓	▲ Improved	15%	0%
case-02	✓→✓	＝ Same ✓	130%	0%
case-03	✓→✓	＝ Same ✓	78%	0%
case-05	✓→✓	＝ Same ✓	50%	0%

Showing 5 of 22 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/3/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

aiskillstore / marketplace

skills/89jobrien/prompt-optimization/SKILL.md

View on GitHub ↗

Imported 8/3/2026Version v1

Skill contents

SKILL.md

Prompt Optimization

This skill optimizes prompts for LLMs and AI systems, focusing on effective prompt patterns, few-shot learning, and optimal AI interactions.

When to Use This Skill

When building AI features or agents
When improving LLM response quality
When crafting system prompts
When optimizing agent performance
When implementing few-shot learning
When designing AI workflows

What This Skill Does

Prompt Design: Creates effective prompts with clear structure
Few-Shot Learning: Implements few-shot examples for better results
Chain-of-Thought: Uses reasoning patterns for complex tasks
Output Formatting: Specifies clear output formats
Constraint Setting: Sets boundaries and constraints
Performance Optimization: Improves prompt efficiency and results

How to Use

Optimize Prompt

Optimize this prompt for better results

Create a system prompt for a code review agent

Specific Patterns

Implement few-shot learning for this task

Prompt Techniques

Structure

Clear Sections:

Role definition
Task description
Constraints and boundaries
Output format
Examples

Few-Shot Learning

Pattern:

Provide 2-3 examples
Show input-output pairs
Demonstrate desired style
Include edge cases

Chain-of-Thought

Approach:

Break down complex tasks
Show reasoning steps
Encourage step-by-step thinking
Verify intermediate results

Examples

Example 1: Code Review Prompt

Input: Create optimized code review prompt

Output:

markdown
## Optimized Prompt: Code Review

### The Prompt

You are an expert code reviewer with 10+ years of experience.

Review the provided code focusing on:

Security vulnerabilities
Performance optimizations
Code maintainability
Best practices

For each issue found, provide:

Severity level (Critical/High/Medium/Low)
Specific line numbers
Explanation of the issue
Suggested fix with code example

Format your response as a structured report with clear sections.


### Techniques Used
- Role-playing for expertise
- Clear evaluation criteria
- Specific output format
- Actionable feedback requirements

Best Practices

Prompt Design

Be Specific: Clear, unambiguous instructions
Provide Examples: Show desired output format
Set Constraints: Define boundaries clearly
Iterate: Test and refine prompts
Document: Keep track of effective patterns

Related Use Cases

AI agent development
LLM optimization
System prompt creation
Few-shot learning implementation
AI workflow design

Want to see if aiskillstore/prompt-optimization would activate on your phrasing? Try it before you install — no LLM call, no signup.

Latest benchmark · publisher runVerified

8/21/2026, 20:05 UTC · 22 cases

Lift

+9%

95% vs 86%

Agent turns

1.0 vs 1.0

Tokens

+44%

2,169.41 vs 1,505.73

fail → pass · 2pass → pass · 19fail → fail · 1

Per-case results (22)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-01 We are configuring an OpenAI API request for an automated entity extraction pipeline that must produce valid JSON adhering to a schema without extra markdown text. A developer suggests relying solely on system prompt instructions like 'Output valid JSON only'. What API parameter should be set instead?	fail→fail	6,709	7,564	+13%	1	1	0%	1,230	1,843	+50%	0	0	—
case-02 We are deploying a prompt to classify high-volume customer support tickets into five categories. A developer suggests setting `temperature: 0.7` to make classifications more creative. What temperature value should be configured for deterministic classification?	pass→pass	5,332	7,654	+44%	1	1	0%	825	1,899	+130%	0	0	—
case-03 When structuring system prompts for Anthropic Claude models to separate user instructions, reference material, and few-shot examples, what structural markup syntax is prescribed by Anthropic guidelines?	pass→pass	7,500	9,168	+22%	1	1	0%	1,202	2,138	+78%	0	0	—
case-04 To prevent a reasoning model from outputting premature conclusions while keeping the final answer clean, inside which XML tags should the model write its intermediate reasoning steps according to Anthropic prompt patterns?	fail→pass	3,357	4,486	+34%	1	1	0%	538	1,273	+137%	0	0	—
case-05 In OpenAI chat completion API requests, where should persistent system instructions, behavioral boundaries, and persona definitions be placed within the message payload?	pass→pass	8,822	10,439	+18%	1	1	0%	1,528	2,298	+50%	0	0	—
case-06 When including few-shot examples in a Multi-turn Chat Completion API request, should the example pairs be concatenated into a single system text string or formatted as distinct API message objects?	pass→pass	10,443	14,489	+39%	1	1	0%	1,666	2,965	+78%	0	0	—
case-07 When drafting instructions for a Retrieval-Augmented Generation (RAG) assistant, how should the model be instructed to respond when the provided search context lacks the answer?	pass→pass	14,410	12,410	-14%	1	1	0%	2,274	2,547	+12%	0	0	—
case-08 When defining prompt boundaries to prevent unwanted model behavior, why are affirmative explicit constraints preferred over sole negative prohibitions like 'do not do X'?	fail→pass	14,146	12,124	-14%	1	1	0%	2,130	2,445	+15%	0	0	—
case-09 A team wants to optimize a production system prompt. An engineer suggests manually checking 3 playground outputs after every edit. What systematic testing methodology should be used instead?	pass→pass	12,171	14,096	+16%	1	1	0%	1,870	2,802	+50%	0	0	—
case-10 In the Anthropic API, how can a prompt developer force the model to start its response immediately with valid JSON without introductory prose?	pass→pass	11,803	9,542	-19%	1	1	0%	1,962	2,257	+15%	0	0	—
case-11 Instead of manually editing prompt wording and hand-selecting few-shot examples for a complex pipeline, what programmatic framework compiles prompts by optimizing instructions against a metric?	pass→pass	12,166	12,721	+5%	1	1	0%	1,810	2,881	+59%	0	0	—
case-12 When defining a custom function for an LLM agent in an API request, where and how must the function parameter types and descriptions be structured?	pass→pass	12,645	11,410	-10%	1	1	0%	2,235	2,592	+16%	0	0	—
case-13 When tuning generation hyperparameters for an LLM, a developer proposes lowering `temperature` to 0.2 and simultaneously lowering `top_p` to 0.1. What parameter tuning best practice applies?	pass→pass	6,206	8,127	+31%	1	1	0%	987	1,889	+91%	0	0	—
case-14 When optimizing latency and API costs using prompt caching for long context documents, where in the API payload must the static context be located?	pass→pass	8,940	9,727	+9%	1	1	0%	1,415	2,196	+55%	0	0	—
case-15 To prevent user-supplied text from executing prompt injection attacks that override system instructions, how should user input be enclosed in the prompt?	pass→pass	12,204	9,772	-20%	1	1	0%	1,871	2,142	+14%	0	0	—
case-16 What exact trigger phrase, established by Kojima et al. (2022), induces zero-shot chain-of-thought reasoning in LLMs without requiring few-shot examples?	pass→pass	2,698	3,641	+35%	1	1	0%	415	1,156	+179%	0	0	—
case-17 In OpenAI chat completions, what API parameter accepts a key-value dictionary mapping token IDs to bias values from -100 to 100 to suppress specific tokens?	pass→pass	3,851	3,394	-12%	1	1	0%	645	1,167	+81%	0	0	—
case-18 According to findings on the 'lost in the middle' phenomenon in context processing, where in a long multi-document prompt should critical reference context be placed for highest retrieval accuracy?	pass→pass	6,489	7,460	+15%	1	1	0%	1,062	1,732	+63%	0	0	—
case-19 To prevent an LLM from repeating identical phrases verbatim during long text generation without penalizing word reuse based on simple presence, which API parameter should be increased?	pass→pass	4,689	5,995	+28%	1	1	0%	681	1,450	+113%	0	0	—
case-20 What learning rate schedule and batch size should be configured when fine-tuning Llama-3 8B using LoRA on a GPU cluster?	pass→pass	16,599	15,221	-8%	1	1	0%	2,816	3,234	+15%	0	0	—
case-21 Which HNSW index parameters in vector databases control search recall versus query latency during graph traversal?	pass→pass	12,951	12,323	-5%	1	1	0%	2,188	2,611	+19%	0	0	—
case-22 When configuring regex pattern matching to detect SQL injection attempts at an API gateway, which pattern detects basic UNION-based payload injections?	pass→pass	11,309	9,808	-13%	1	1	0%	1,776	2,210	+24%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +9 percentage points is the difference between those two pass rates over the 22 comparable cases.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 86.4%
Pass rate with the skill: 95.5%
Cases attempted: 22
Cases counted toward lift: 22
Cases that regressed: 0
Skill version measured: v1
Run completed: 8/21/2026, 20:05 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-04

Prompt

To prevent a reasoning model from outputting premature conclusions while keeping the final answer clean, inside which XML tags should the model write its intermediate reasoning steps according to Anthropic prompt patterns?

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals1 of 4 complete

BenchmarkPass rate WITH skill on a verified run95

Live evalNeeds ≥20 evaluated traces in 30 days

AI ratingNeeds ≥10 rated traces

AdoptionNeeds ≥10 serves across ≥2 orgs in 30 days

Efficiency · separate axis (at equal correctness)+0% turns+44% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 1 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 7 days ago

Other measured skills in the registry, with their headline benchmark lift.