Koog 1.0 idioms, gotchas, and scaffolding skills for Kotlin agents on the JVM
88
88%
Does it follow best practices?
Impact
88%
1.95xAverage score across 43 eval scenarios
Passed
No known issues
{
"context": "Negative case — tests whether the agent identifies the developer's situation (stable system prompt content with unique per-call user inputs) as a provider-side caching problem, not a response caching problem, and routes them to the right tile skill. Wrapping the executor in an in-process response cache here would be wrong: responses are not actually repeated, so the cache would never hit and the developer would conclude caching doesn't work.",
"type": "weighted_checklist",
"checklist": [
{
"name": "Names the cache-vs-response distinction",
"description": "Surfaces that there are two different caching layers — provider-side reduction on stable prompt prefixes vs. in-process avoidance of duplicate API calls — and identifies which one this developer needs. Without this distinction the developer cannot decide between offerings",
"max_score": 25
},
{
"name": "Recommends the provider-side caching path",
"description": "Recommends the Anthropic prompt caching surface (cacheControl breakpoints on the system content) — or invokes the appropriate sibling skill. Does not invent a custom cache header injection and does not point at a generic OpenTelemetry feature",
"max_score": 30
},
{
"name": "Does not wrap the executor in an in-process cache",
"description": "Does NOT produce a CachedPromptExecutor / FilePromptCache / RedisPromptCache wrapper around the developer's executor. The user inputs are unique per call so the cache key would never hit — implementing this would mislead the developer into thinking caching is broken",
"max_score": 25
},
{
"name": "Acknowledges the framing without capitulating",
"description": "Engages with the developer's phrasing (\"cache the LLM responses\") and explains that what they actually want bills the API at a reduced rate rather than skipping the API. Does not just refuse without naming the mismatch",
"max_score": 10
},
{
"name": "Mentions Anthropic's minimum-token requirement",
"description": "Notes that Anthropic enforces a minimum token threshold for caching to take effect — the developer's multi-thousand-word prompt likely qualifies, but breakpoints on shorter content are silently ignored. They need this so a no-savings outcome isn't a surprise",
"max_score": 10
}
]
}evals
scenario-1
scenario-2
scenario-3
scenario-4
scenario-5
scenario-6
scenario-7
scenario-8
scenario-9
scenario-10
scenario-11
scenario-12
scenario-13
scenario-14
scenario-15
scenario-16
scenario-17
scenario-18
scenario-19
scenario-20
scenario-21
scenario-22
scenario-23
scenario-24
scenario-25
scenario-26
scenario-27
scenario-28
scenario-29
scenario-30
scenario-31
scenario-32
scenario-33
scenario-34
scenario-35
scenario-36
scenario-37
scenario-38
scenario-39
scenario-40
scenario-41
scenario-42
scenario-43
skills
add-observability
add-persistence
add-rag
add-structured-output
add-token-budgeting
add-tool
cache-llm-calls
define-prompt
domain-model-subtask-pipeline
references
enable-prompt-caching
handle-agent-events
manage-state
migrate-from-0-x
model-planner-subtasks
persist-chat-history
query-sql-from-agent
scaffold-agent
snapshot-and-restore
test-koog-agents
trace-agent-internals
use-attachments
use-functional-agent
use-llm-node-variants
use-planner
wire-a2a
wire-acp-server
wire-ktor-server
wire-mcp-server
wire-spring-boot