agent-evaluation

Name: agent-evaluation
Author: NeoLabHQ

877installs132forks

How to install agent-evaluation?

neolabhq/context-engineering-kit/agent-evaluation

$npx -y skills add NeoLabHQ/context-engineering-kit --skill agent-evaluation

Installs into the current project.

Summary

Evaluate and improve Claude Code commands, skills, and agents. Use when testing prompt effectiveness, validating context engineering choices, or measuring improvement quality.

Files · 1

View on GitHub

SKILL.md

1	# Evaluation Methods for Claude Code Agents
2
3	Evaluation of agent systems requires different approaches than traditional software or even standard language model applications. Agents make dynamic decisions, are non-deterministic between runs, and often lack single correct answers. Effective evaluation must account for these characteristics while providing actionable feedback. A robust evaluation framework enables continuous improvement, catches regressions, and validates that context engineering choices achieve intended effects.
4
5	## Core Concepts
6
7	Agent evaluation requires outcome-focused approaches that account for non-determinism and multiple valid paths. Multi-dimensional rubrics capture various quality aspects: factual accuracy, completeness, citation accuracy, source quality, and tool efficiency. LLM-as-judge provides scalable evaluation while human evaluation catches edge cases.
8
9	The key insight is that agents may find alternative paths to goals—the evaluation should judge whether they achieve right outcomes while following reasonable processes.
10
11	Performance Drivers: The 95% Finding
12	Research on the BrowseComp evaluation (which tests browsing agents' ability to locate hard-to-find information) found that three factors explain 95% of performance variance:
13
14	\| Factor \| Variance Explained \| Implication \|
15	\|--------\|-------------------\|-------------\|
16	\| Token usage \| 80% \| More tokens = better performance \|
17	\| Number of tool calls \| ~10% \| More exploration helps \|
18	\| Model choice \| ~5% \| Better models multiply efficiency \|
19
20	Implications for Claude Code development:
21
22	- Token budgets matter: Evaluate with realistic token constraints
23	- Model upgrades beat token increases: Upgrading models provides larger gains than increasing token budgets
24	- Multi-agent validation: Validates architectures that distribute work across subagents with separate context windows
25
26	## Evaluation Challenges
27
28	### Non-Determinism and Multiple Valid Paths
29
30	Agents may take completely different valid paths to reach goals. One agent might search three sources while another searches ten. They might use different tools to find the same answer. Traditional evaluations that check for specific steps fail in this context.
31
32	Solution: The solution is outcomes, not exact execution paths. Judge whether the agent achieves the right result through a reasonable process.
33
34	### Context-Dependent Failures
35
36	Agent failures often depend on context in subtle ways. An agent might succeed on complex queries but fail on simple ones. It might work well with one tool set but fail with another. Failures may emerge only after extended interaction when context accumulates.
37
38	Solution: Evaluation must cover a range of complexity levels and test extended interactions, not just isolated queries.
39
40	### Composite Quality Dimensions
41
42	Agent quality is not a single dimension. It includes factual accuracy, completeness, coherence, tool efficiency, and process quality. An agent might score high on accuracy but low in efficiency, or vice versa.
43
44	An agent might score high on accuracy but low in efficiency.
45
46	Solution: Evaluation rubrics must capture multiple dimensions with appropriate weighting for the use case.
47
48	## Evaluation Rubric Design
49
50	### Multi-Dimensional Rubric
51
52	Effective rubrics cover key dimensions with descriptive levels:
53
54	Instruction Following (weight: 0.30)
55
56	- Excellent (1.0): All instructions followed precisely
57	- Good (0.8): Minor deviations that don't affect outcome
58	- Acceptable (0.6): Major instructions followed, minor ones missed
59	- Poor (0.3): Significant instructions ignored
60	- Failed (0.0): Fundamentally misunderstood the task
61
62	Output Completeness (weight: 0.25)
63
64	- Excellent: All requested aspects thoroughly covered
65	- Good: Most aspects covered with minor gaps
66	- Acceptable: Key aspects covered, some gaps
67	- Poor: Major aspects missing
68	- Failed: Fundamental aspects not addressed
69
70	Tool Efficiency (weight: 0.20)
71
72	- Excellent: Optimal tool selection and minimal calls
73	- Good: Good tool selection with minor inefficiencies
74	- Acceptable: Appropriate tools with some redundancy
75	- Poor: Wrong tools or excessive calls
76	- Failed: Severe tool misuse or extremely excessive calls
77
78	Reasoning Quality (weight: 0.15)
79
80	- Excellent: Clear, logical reasoning throughout
81	- Good: Generally sound reasoning with minor gaps
82	- Acceptable: Basic reasoning present
83	- Poor: Reasoning unclear or flawed
84	- Failed: No apparent reasoning
85
86	Response Coherence (weight: 0.10)
87
88	- Excellent: Well-structured, easy to follow
89	- Good: Generally coherent with minor issues
90	- Acceptable: Understandable but could be clearer
91	- Poor: Difficult to follow
92	- Failed: Incoherent
93
94	### Scoring Approach
95
96	Convert dimension assessments to numeric scores (0.0 to 1.0) with a

People who liked this also viewed

6 picks

Type

agent-evaluation

877installs132forks

View on GitHub

How to install agent-evaluation?

neolabhq/context-engineering-kit/agent-evaluation

$npx -y skills add NeoLabHQ/context-engineering-kit --skill agent-evaluation

Installs into the current project.

Summary

Evaluate and improve Claude Code commands, skills, and agents. Use when testing prompt effectiveness, validating context engineering choices, or measuring improvement quality.

Files · 1

View on GitHub

SKILL.md

1	# Evaluation Methods for Claude Code Agents
2
3	Evaluation of agent systems requires different approaches than traditional software or even standard language model applications. Agents make dynamic decisions, are non-deterministic between runs, and often lack single correct answers. Effective evaluation must account for these characteristics while providing actionable feedback. A robust evaluation framework enables continuous improvement, catches regressions, and validates that context engineering choices achieve intended effects.
4
5	## Core Concepts
6
7	Agent evaluation requires outcome-focused approaches that account for non-determinism and multiple valid paths. Multi-dimensional rubrics capture various quality aspects: factual accuracy, completeness, citation accuracy, source quality, and tool efficiency. LLM-as-judge provides scalable evaluation while human evaluation catches edge cases.
8
9	The key insight is that agents may find alternative paths to goals—the evaluation should judge whether they achieve right outcomes while following reasonable processes.
10
11	Performance Drivers: The 95% Finding
12	Research on the BrowseComp evaluation (which tests browsing agents' ability to locate hard-to-find information) found that three factors explain 95% of performance variance:
13
14	\| Factor \| Variance Explained \| Implication \|
15	\|--------\|-------------------\|-------------\|
16	\| Token usage \| 80% \| More tokens = better performance \|
17	\| Number of tool calls \| ~10% \| More exploration helps \|
18	\| Model choice \| ~5% \| Better models multiply efficiency \|
19
20	Implications for Claude Code development:
21
22	- Token budgets matter: Evaluate with realistic token constraints
23	- Model upgrades beat token increases: Upgrading models provides larger gains than increasing token budgets
24	- Multi-agent validation: Validates architectures that distribute work across subagents with separate context windows
25
26	## Evaluation Challenges
27
28	### Non-Determinism and Multiple Valid Paths
29
30	Agents may take completely different valid paths to reach goals. One agent might search three sources while another searches ten. They might use different tools to find the same answer. Traditional evaluations that check for specific steps fail in this context.
31
32	Solution: The solution is outcomes, not exact execution paths. Judge whether the agent achieves the right result through a reasonable process.
33
34	### Context-Dependent Failures
35
36	Agent failures often depend on context in subtle ways. An agent might succeed on complex queries but fail on simple ones. It might work well with one tool set but fail with another. Failures may emerge only after extended interaction when context accumulates.
37
38	Solution: Evaluation must cover a range of complexity levels and test extended interactions, not just isolated queries.
39
40	### Composite Quality Dimensions
41
42	Agent quality is not a single dimension. It includes factual accuracy, completeness, coherence, tool efficiency, and process quality. An agent might score high on accuracy but low in efficiency, or vice versa.
43
44	An agent might score high on accuracy but low in efficiency.
45
46	Solution: Evaluation rubrics must capture multiple dimensions with appropriate weighting for the use case.
47
48	## Evaluation Rubric Design
49
50	### Multi-Dimensional Rubric
51
52	Effective rubrics cover key dimensions with descriptive levels:
53
54	Instruction Following (weight: 0.30)
55
56	- Excellent (1.0): All instructions followed precisely
57	- Good (0.8): Minor deviations that don't affect outcome
58	- Acceptable (0.6): Major instructions followed, minor ones missed
59	- Poor (0.3): Significant instructions ignored
60	- Failed (0.0): Fundamentally misunderstood the task
61
62	Output Completeness (weight: 0.25)
63
64	- Excellent: All requested aspects thoroughly covered
65	- Good: Most aspects covered with minor gaps
66	- Acceptable: Key aspects covered, some gaps
67	- Poor: Major aspects missing
68	- Failed: Fundamental aspects not addressed
69
70	Tool Efficiency (weight: 0.20)
71
72	- Excellent: Optimal tool selection and minimal calls
73	- Good: Good tool selection with minor inefficiencies
74	- Acceptable: Appropriate tools with some redundancy
75	- Poor: Wrong tools or excessive calls
76	- Failed: Severe tool misuse or extremely excessive calls
77
78	Reasoning Quality (weight: 0.15)
79
80	- Excellent: Clear, logical reasoning throughout
81	- Good: Generally sound reasoning with minor gaps
82	- Acceptable: Basic reasoning present
83	- Poor: Reasoning unclear or flawed
84	- Failed: No apparent reasoning
85
86	Response Coherence (weight: 0.10)
87
88	- Excellent: Well-structured, easy to follow
89	- Good: Generally coherent with minor issues
90	- Acceptable: Understandable but could be clearer
91	- Poor: Difficult to follow
92	- Failed: Incoherent
93
94	### Scoring Approach
95
96	Convert dimension assessments to numeric scores (0.0 to 1.0) with a

People who liked this also viewed

6 picks

Type