evaluation

Name: evaluation
Author: muratcankoylan

52installs1.4kforks

How to install evaluation?

muratcankoylan/agent-skills-for-context-engineering/evaluation

$npx -y skills add muratcankoylan/Agent-Skills-for-Context-Engineering --skill evaluation

Installs into the current project.

Summary

This skill should be used when building agent evaluation systems: deterministic checks, regression suites, multi-dimensional rubrics, quality gates, production monitoring, baseline comparison, and outcome measurement for agent pipelines.

Files · 1

View on GitHub

SKILL.md

1	# Evaluation Methods for Agent Systems
2
3	Evaluate agent systems differently from traditional software because agents make dynamic decisions, are non-deterministic between runs, and often lack single correct answers. Build evaluation frameworks that account for these characteristics, provide actionable feedback, catch regressions, and validate that context engineering choices achieve intended effects.
4
5	## When to Activate
6
7	Activate this skill when:
8	- Testing agent performance systematically
9	- Validating context engineering choices
10	- Measuring improvements over time
11	- Catching regressions before deployment
12	- Building quality gates for agent pipelines
13	- Comparing different agent configurations
14	- Evaluating production systems continuously
15
16	Do not activate this skill for adjacent work owned by other skills:
17	- Designing the LLM judge itself, pairwise comparison, judge calibration, or bias mitigation: `advanced-evaluation`.
18	- Designing autonomous control surfaces, novelty gates, rollback, or PR approval boundaries: `harness-engineering`.
19	- Debugging a specific context failure mode before measuring it: `context-degradation`.
20
21	## Core Concepts
22
23	Focus evaluation on outcomes rather than execution paths, because agents may find alternative valid routes to goals. Judge whether the agent achieves the right outcome via a reasonable process, not whether it followed a specific sequence of steps.
24
25	Use multi-dimensional rubrics instead of single scores because one number hides critical failures in specific dimensions. Capture factual accuracy, completeness, citation accuracy, source quality, and tool efficiency as separate dimensions, then weight them for the use case.
26
27	Use model-judged evaluation only after deterministic checks and rubrics are stable. When the work centers on judge prompts, pairwise comparison, calibration, or bias mitigation, switch to Advanced Evaluation.
28
29	Run deterministic validation before LLM judgment whenever the artifact has machine-checkable structure. Schema validity, duplicate keys, rubric math, manifest sync, retrieval status, and required evidence paths should fail fast before an evaluator spends tokens or returns a subjective score.
30
31	Performance Drivers
32
33	Apply browsing-agent research when designing evaluation budgets: token usage, tool calls, and model choice can dominate measured performance variance (claim-evaluation-browsecomp-variance).
34
35	\| Factor \| Variance Explained \| Implication \|
36	\|--------\|-------------------\|-------------\|
37	\| Token usage \| Primary driver \| More exploration can improve performance until cost or context quality collapses \|
38	\| Number of tool calls \| Secondary driver \| More tool use helps only when calls retrieve useful evidence \|
39	\| Model choice \| Secondary but multiplicative \| Better models often use tokens and tools more efficiently \|
40
41	Act on these implications when designing evaluations:
42	- Set realistic token budgets: Evaluate agents with production-realistic token limits, not unlimited resources.
43	- Compare model upgrades against token increases: Better models may use tokens more efficiently than weaker models with larger budgets.
44	- Validate multi-agent architectures: Extra agents add tokens and tool calls; evaluate them against single-agent baselines.
45
46	## Detailed Topics
47
48	### Evaluation Challenges
49
50	Handle Non-Determinism and Multiple Valid Paths
51
52	Design evaluations that tolerate path variation because agents may take completely different valid paths to reach goals. One agent might search three sources while another searches ten; both may produce correct answers. Avoid checking for specific steps. Instead, define outcome criteria (correctness, completeness, quality) and score against those, treating the execution path as informational rather than evaluative.
53
54	Test Context-Dependent Failures
55
56	Evaluate across a range of complexity levels and interaction lengths because agent failures often depend on context in subtle ways. An agent might succeed on simple queries but fail on complex ones, work well with one tool set but fail with another, or degrade after extended interaction as context accumulates. Include simple, medium, complex, and very complex test cases to surface these patterns.
57
58	Score Composite Quality Dimensions Separately
59
60	Break agent quality into separate dimensions (factual accuracy, completeness, coherence, tool efficiency, process quality) and score each independently because an agent might score high on accuracy but low on efficiency, or vice versa. Then compute weighted aggregates tuned to use-case priorities. This approach reveals which dimensions need improvement rather than averaging away the signal.
61
62	### Evaluation Rubric Design
63
64	**Build Multi-Dimensional

People who liked this also viewed

6 picks

Type

evaluation

52installs1.4kforks

View on GitHub

How to install evaluation?

muratcankoylan/agent-skills-for-context-engineering/evaluation

$npx -y skills add muratcankoylan/Agent-Skills-for-Context-Engineering --skill evaluation

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Evaluation Methods for Agent Systems
2
3	Evaluate agent systems differently from traditional software because agents make dynamic decisions, are non-deterministic between runs, and often lack single correct answers. Build evaluation frameworks that account for these characteristics, provide actionable feedback, catch regressions, and validate that context engineering choices achieve intended effects.
4
5	## When to Activate
6
7	Activate this skill when:
8	- Testing agent performance systematically
9	- Validating context engineering choices
10	- Measuring improvements over time
11	- Catching regressions before deployment
12	- Building quality gates for agent pipelines
13	- Comparing different agent configurations
14	- Evaluating production systems continuously
15
16	Do not activate this skill for adjacent work owned by other skills:
17	- Designing the LLM judge itself, pairwise comparison, judge calibration, or bias mitigation: `advanced-evaluation`.
18	- Designing autonomous control surfaces, novelty gates, rollback, or PR approval boundaries: `harness-engineering`.
19	- Debugging a specific context failure mode before measuring it: `context-degradation`.
20
21	## Core Concepts
22
23	Focus evaluation on outcomes rather than execution paths, because agents may find alternative valid routes to goals. Judge whether the agent achieves the right outcome via a reasonable process, not whether it followed a specific sequence of steps.
24
25	Use multi-dimensional rubrics instead of single scores because one number hides critical failures in specific dimensions. Capture factual accuracy, completeness, citation accuracy, source quality, and tool efficiency as separate dimensions, then weight them for the use case.
26
27	Use model-judged evaluation only after deterministic checks and rubrics are stable. When the work centers on judge prompts, pairwise comparison, calibration, or bias mitigation, switch to Advanced Evaluation.
28
29	Run deterministic validation before LLM judgment whenever the artifact has machine-checkable structure. Schema validity, duplicate keys, rubric math, manifest sync, retrieval status, and required evidence paths should fail fast before an evaluator spends tokens or returns a subjective score.
30
31	Performance Drivers
32
33	Apply browsing-agent research when designing evaluation budgets: token usage, tool calls, and model choice can dominate measured performance variance (claim-evaluation-browsecomp-variance).
34
35	\| Factor \| Variance Explained \| Implication \|
36	\|--------\|-------------------\|-------------\|
37	\| Token usage \| Primary driver \| More exploration can improve performance until cost or context quality collapses \|
38	\| Number of tool calls \| Secondary driver \| More tool use helps only when calls retrieve useful evidence \|
39	\| Model choice \| Secondary but multiplicative \| Better models often use tokens and tools more efficiently \|
40
41	Act on these implications when designing evaluations:
42	- Set realistic token budgets: Evaluate agents with production-realistic token limits, not unlimited resources.
43	- Compare model upgrades against token increases: Better models may use tokens more efficiently than weaker models with larger budgets.
44	- Validate multi-agent architectures: Extra agents add tokens and tool calls; evaluate them against single-agent baselines.
45
46	## Detailed Topics
47
48	### Evaluation Challenges
49
50	Handle Non-Determinism and Multiple Valid Paths
51
52	Design evaluations that tolerate path variation because agents may take completely different valid paths to reach goals. One agent might search three sources while another searches ten; both may produce correct answers. Avoid checking for specific steps. Instead, define outcome criteria (correctness, completeness, quality) and score against those, treating the execution path as informational rather than evaluative.
53
54	Test Context-Dependent Failures
55
56	Evaluate across a range of complexity levels and interaction lengths because agent failures often depend on context in subtle ways. An agent might succeed on simple queries but fail on complex ones, work well with one tool set but fail with another, or degrade after extended interaction as context accumulates. Include simple, medium, complex, and very complex test cases to surface these patterns.
57
58	Score Composite Quality Dimensions Separately
59
60	Break agent quality into separate dimensions (factual accuracy, completeness, coherence, tool efficiency, process quality) and score each independently because an agent might score high on accuracy but low on efficiency, or vice versa. Then compute weighted aggregates tuned to use-case priorities. This approach reveals which dimensions need improvement rather than averaging away the signal.
61
62	### Evaluation Rubric Design
63
64	**Build Multi-Dimensional

People who liked this also viewed

6 picks

Type