CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

online-evals

Attach judges to config variations for automatic LLM-as-a-judge evaluation. Create custom judges, configure sampling rates, and monitor quality scores.

Quality

75%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./skills/agentcontrol/online-evals/SKILL.md
SKILL.md
Quality
Evals
Security

Quality

Content

71%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

The body delivers highly actionable, executable guidance across MCP, REST, and SDK paths with a clear three-step workflow, explicit gotchas (judge array replacement, {{message_history}} requirement), and an error-handling table. Its weaknesses are structural: everything is inlined in one long file with no progressive disclosure to bundle files, and a couple of code snippets have minor executability gaps.

Suggestions

Move the REST client class to scripts/ (e.g., scripts/attach_judges.py) and the inline/background/pipeline SDK examples to references/ (e.g., references/sdk-python.md, references/sdk-node.md), keeping SKILL.md as a workflow overview with one representative example per path.

Add an explicit verification step after Step 2 (e.g., re-call get-agent-configs to confirm the judgeConfiguration was applied) to close the workflow validation gap.

Fix the background-evaluation Python snippet by wrapping the invoke/run_judge calls in an async function so the code is executable as written.

DimensionReasoningScore

Conciseness

The body is dense and operational with almost no explanation of concepts Claude already knows, but the ~140-line inlined REST manager class, six dual-language SDK code blocks, and version pins ("0.3.0+") are minor over-explanations that could be trimmed. Not 5 because those sections could be tightened or offloaded; not 3 because there is no concept padding of the kind the lower anchors describe.

4 / 5

Actionability

MCP JSON payloads are copy-paste ready with placeholders, and the REST class is complete with error handling. Not 5 because the background-evaluation Python snippet uses top-level "await" with no async wrapper, and the custom-pipeline example leaves "user_input" and "context" undefined (only partially justified by the "run_my_agent is your existing pipeline" note).

4 / 5

Workflow Clarity

Steps 1-3 are clearly sequenced with an explicit validation checkpoint ("call get-agent-configs with the judge's configKey and section: \"health\" to confirm it's targetable") and an error-recovery table mapping status codes to solutions. Not 5 because there is no explicit verification step after the Step 2 attachment itself, and no feedback loop for retrying failed attachments.

4 / 5

Progressive Disclosure

Section headers are clear and Related Skills/References are well signaled, but no bundle files exist: the full REST client class clearly belongs in scripts/ and the six SDK examples belong in references/, yet all are inlined in a ~460-line SKILL.md. This matches anchor 3 (content that should be separate is inline) rather than 4, since the bulk of the operational detail lives in the main file.

3 / 5

Total

15

/

20

Passed

Description

66%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

The description is specific, action-oriented, and written in third person, listing four concrete capabilities with good natural keywords. Its main weakness is the absence of any "Use when..." trigger clause, which both caps completeness and limits trigger-term coverage for users who need this skill.

Suggestions

Add an explicit trigger clause such as "Use when setting up LLM-as-a-judge evaluation for LaunchDarkly AI configs, or when the user mentions judges, evaluation metrics, or quality scoring for config variations."

Include one or two natural synonyms (e.g., "scoring", "AI config evaluation") alongside "LLM-as-a-judge evaluation" to broaden trigger-term coverage.

Mention SDK-side evaluation (running judges inline or in the background from Python/Node.js) so the description covers the full capability set the body delivers.

DimensionReasoningScore

Specificity

"Attach judges to config variations", "Create custom judges", "configure sampling rates", and "monitor quality scores" are four concrete, domain-anchored actions. Not 5 because SDK-side capabilities (inline/background evaluation, guardrail integration) described in the body are unmentioned, leaving minor coverage gaps.

4 / 5

Completeness

The "what" is clear and multi-part (attach judges, create custom judges, configure sampling, monitor scores), but there is no "Use when..." clause or equivalent explicit trigger guidance, which caps completeness at 3 per the judging guidelines.

3 / 5

Trigger Term Quality

Natural user phrases like "judges", "LLM-as-a-judge", "evaluation", "quality scores", and "sampling rates" are present. Not 5 because common synonyms such as "scoring", "LLM evaluation", or "AI config quality" are missing.

4 / 5

Distinctiveness Conflict Risk

The LaunchDarkly-specific niche ("config variations", "judges", LLM-as-a-judge evaluation) is mostly distinct. Not 5 because "config variations" overlaps with sibling skills like configs-create and configs-targeting.

4 / 5

Total

15

/

20

Passed

Validation

100%

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository

launchdarkly/ai-tooling

Path

skills/agentcontrol/online-evals/SKILL.md

Reviewed

Table of Contents

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.