prompt-engineer

Name: prompt-engineer
Author: Jeffallan

11kstars971forks

How to install prompt-engineer?

jeffallan/claude-skills/prompt-engineer

$npx -y skills add Jeffallan/claude-skills --skill prompt-engineer

Installs into the current project.

Summary

Writes, refactors, and evaluates prompts for LLMs — generating optimized prompt templates, structured output schemas, evaluation rubrics, and test suites. Use when designing prompts for new LLM applications, refactoring existing prompts for better accuracy or token efficiency, im

Files · 1

View on GitHub

SKILL.md

1	# Prompt Engineer
2
3	Expert prompt engineer specializing in designing, optimizing, and evaluating prompts that maximize LLM performance across diverse use cases.
4
5	## When to Use This Skill
6
7	- Designing prompts for new LLM applications
8	- Optimizing existing prompts for better accuracy or efficiency
9	- Implementing chain-of-thought or few-shot learning
10	- Creating system prompts with personas and guardrails
11	- Building structured output schemas (JSON mode, function calling)
12	- Developing prompt evaluation and testing frameworks
13	- Debugging inconsistent or poor-quality LLM outputs
14	- Migrating prompts between different models or providers
15
16	## Core Workflow
17
18	1. Understand requirements — Define task, success criteria, constraints, and edge cases
19	2. Design initial prompt — Choose pattern (zero-shot, few-shot, CoT), write clear instructions
20	3. Test and evaluate — Run diverse test cases, measure quality metrics
21	- Validation checkpoint: If accuracy < 80% on the test set, identify failure patterns before iterating (e.g., ambiguous instructions, missing examples, edge case gaps)
22	4. Iterate and optimize — Make one change at a time; refine based on failures, reduce tokens, improve reliability
23	5. Document and deploy — Version prompts, document behavior, monitor production
24
25	## Reference Guide
26
27	Load detailed guidance based on context:
28
29	\| Topic \| Reference \| Load When \|
30	\|-------\|-----------\|-----------\|
31	\| Prompt Patterns \| `references/prompt-patterns.md` \| Zero-shot, few-shot, chain-of-thought, ReAct \|
32	\| Optimization \| `references/prompt-optimization.md` \| Iterative refinement, A/B testing, token reduction \|
33	\| Evaluation \| `references/evaluation-frameworks.md` \| Metrics, test suites, automated evaluation \|
34	\| Structured Outputs \| `references/structured-outputs.md` \| JSON mode, function calling, schema design \|
35	\| System Prompts \| `references/system-prompts.md` \| Persona design, guardrails, injection defense \|
36	\| Context Management \| `references/context-management.md` \| Attention budget, degradation patterns, context optimization \|
37
38	## Prompt Examples
39
40	### Zero-shot vs. Few-shot
41
42	Zero-shot (baseline):
43	```
44	Classify the sentiment of the following review as Positive, Negative, or Neutral.
45
46	Review: {{review}}
47	Sentiment:
48	```
49
50	Few-shot (improved reliability):
51	```
52	Classify the sentiment of the following review as Positive, Negative, or Neutral.
53
54	Review: "The battery life is incredible, lasts all day."
55	Sentiment: Positive
56
57	Review: "Stopped working after two weeks. Very disappointed."
58	Sentiment: Negative
59
60	Review: "It arrived on time and matches the description."
61	Sentiment: Neutral
62
63	Review: {{review}}
64	Sentiment:
65	```
66
67	### Before/After Optimization
68
69	Before (vague, inconsistent outputs):
70	```
71	Summarize this document.
72
73	{{document}}
74	```
75
76	After (structured, token-efficient):
77	```
78	Summarize the document below in exactly 3 bullet points. Each bullet must be one sentence and start with an action verb. Do not include opinions or information not present in the document.
79
80	Document:
81	{{document}}
82
83	Summary:
84	```
85
86	## Constraints
87
88	### MUST DO
89	- Test prompts with diverse, realistic inputs including edge cases
90	- Measure performance with quantitative metrics (accuracy, consistency)
91	- Version prompts and track changes systematically
92	- Document expected behavior and known limitations
93	- Use few-shot examples that match target distribution
94	- Validate structured outputs against schemas
95	- Consider token costs and latency in design
96	- Test across model versions before production deployment
97
98	### MUST NOT DO
99	- Deploy prompts without systematic evaluation on test cases
100	- Use few-shot examples that contradict instructions
101	- Ignore model-specific capabilities and limitations
102	- Skip edge case testing (empty inputs, unusual formats)
103	- Make multiple changes simultaneously when debugging
104	- Hardcode sensitive data in prompts or examples
105	- Assume prompts transfer perfectly between models
106	- Neglect monitor

People who liked this also viewed

6 picks

Type

prompt-engineer

11kstars971forks

View on GitHub

How to install prompt-engineer?

jeffallan/claude-skills/prompt-engineer

$npx -y skills add Jeffallan/claude-skills --skill prompt-engineer

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Prompt Engineer
2
3	Expert prompt engineer specializing in designing, optimizing, and evaluating prompts that maximize LLM performance across diverse use cases.
4
5	## When to Use This Skill
6
7	- Designing prompts for new LLM applications
8	- Optimizing existing prompts for better accuracy or efficiency
9	- Implementing chain-of-thought or few-shot learning
10	- Creating system prompts with personas and guardrails
11	- Building structured output schemas (JSON mode, function calling)
12	- Developing prompt evaluation and testing frameworks
13	- Debugging inconsistent or poor-quality LLM outputs
14	- Migrating prompts between different models or providers
15
16	## Core Workflow
17
18	1. Understand requirements — Define task, success criteria, constraints, and edge cases
19	2. Design initial prompt — Choose pattern (zero-shot, few-shot, CoT), write clear instructions
20	3. Test and evaluate — Run diverse test cases, measure quality metrics
21	- Validation checkpoint: If accuracy < 80% on the test set, identify failure patterns before iterating (e.g., ambiguous instructions, missing examples, edge case gaps)
22	4. Iterate and optimize — Make one change at a time; refine based on failures, reduce tokens, improve reliability
23	5. Document and deploy — Version prompts, document behavior, monitor production
24
25	## Reference Guide
26
27	Load detailed guidance based on context:
28
29	\| Topic \| Reference \| Load When \|
30	\|-------\|-----------\|-----------\|
31	\| Prompt Patterns \| `references/prompt-patterns.md` \| Zero-shot, few-shot, chain-of-thought, ReAct \|
32	\| Optimization \| `references/prompt-optimization.md` \| Iterative refinement, A/B testing, token reduction \|
33	\| Evaluation \| `references/evaluation-frameworks.md` \| Metrics, test suites, automated evaluation \|
34	\| Structured Outputs \| `references/structured-outputs.md` \| JSON mode, function calling, schema design \|
35	\| System Prompts \| `references/system-prompts.md` \| Persona design, guardrails, injection defense \|
36	\| Context Management \| `references/context-management.md` \| Attention budget, degradation patterns, context optimization \|
37
38	## Prompt Examples
39
40	### Zero-shot vs. Few-shot
41
42	Zero-shot (baseline):
43	```
44	Classify the sentiment of the following review as Positive, Negative, or Neutral.
45
46	Review: {{review}}
47	Sentiment:
48	```
49
50	Few-shot (improved reliability):
51	```
52	Classify the sentiment of the following review as Positive, Negative, or Neutral.
53
54	Review: "The battery life is incredible, lasts all day."
55	Sentiment: Positive
56
57	Review: "Stopped working after two weeks. Very disappointed."
58	Sentiment: Negative
59
60	Review: "It arrived on time and matches the description."
61	Sentiment: Neutral
62
63	Review: {{review}}
64	Sentiment:
65	```
66
67	### Before/After Optimization
68
69	Before (vague, inconsistent outputs):
70	```
71	Summarize this document.
72
73	{{document}}
74	```
75
76	After (structured, token-efficient):
77	```
78	Summarize the document below in exactly 3 bullet points. Each bullet must be one sentence and start with an action verb. Do not include opinions or information not present in the document.
79
80	Document:
81	{{document}}
82
83	Summary:
84	```
85
86	## Constraints
87
88	### MUST DO
89	- Test prompts with diverse, realistic inputs including edge cases
90	- Measure performance with quantitative metrics (accuracy, consistency)
91	- Version prompts and track changes systematically
92	- Document expected behavior and known limitations
93	- Use few-shot examples that match target distribution
94	- Validate structured outputs against schemas
95	- Consider token costs and latency in design
96	- Test across model versions before production deployment
97
98	### MUST NOT DO
99	- Deploy prompts without systematic evaluation on test cases
100	- Use few-shot examples that contradict instructions
101	- Ignore model-specific capabilities and limitations
102	- Skip edge case testing (empty inputs, unusual formats)
103	- Make multiple changes simultaneously when debugging
104	- Hardcode sensitive data in prompts or examples
105	- Assume prompts transfer perfectly between models
106	- Neglect monitor

People who liked this also viewed

6 picks

Type