CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

launchdarkly-experiment-hypothesis-builder

Help a user turn a rough idea into a strong, testable experiment hypothesis, or grade one they already wrote. Scores the draft against the Change / Measurement / Rationale rubric, scaffolds an If/then/because sentence with fill-in questions for what is missing, and detects junk input, A/A tests, and multiple measurements. Coaches in conversation, or returns the experiment builder's JSON contract when the caller asks for JSON only. Can check existing flags and metrics with read-only lookups, then hands off to launchdarkly-experiment-setup. Never writes to LaunchDarkly.

Quality

82%

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

SKILL.md
Quality
Evals
Security

Quality

Content

81%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.

An exceptionally actionable, well-sequenced instruction skill: fixed strings, exact JSON contracts, a checked routing order, and explicit feedback loops cover every case including security and fallbacks. Its weaknesses are token efficiency — a statistics backstory, dated provenance comment, and changelog add non-operational weight — and a monolithic single-file layout with no offloading of that non-operational material.

Suggestions

Move or cut the non-operational material: drop the "Where these rules come from" statistics section and the dated sync/provenance comment from the body (keep a one-line pointer if needed), retaining at most a brief changelog.

If version history and the upstream-sync provenance must be kept, move them into a CHANGELOG.md or meta section outside SKILL.md so the loaded context contains only runtime rules.

Tighten the Foundational rules / rubric sections, which repeat the same examples ("grow the business", "users trust familiar payment options") across multiple sections; consolidate to one statement each.

DimensionReasoningScore

Conciseness

The operational core (routing order, hole rules, fixed strings, nudge table, JSON contracts) is lean and earns its tokens, but there is real non-operational padding: the "Where these rules come from" section of 1,171-hypothesis statistics, the dated sync/provenance comment ("@ bcb2e82c (2026-07-31)"), and a two-version changelog with ticket/PR numbers ("EXPT-4665", "#160", "#95") — time-sensitive detail that sits outside any old-patterns/deprecated section. This matches 'mostly efficient but includes some unnecessary explanation or could be tightened'; not 4, because the backstory and provenance material is more than minor over-explanation.

3 / 5

Actionability

As an instruction-only skill the guidance is fully executable: exact fixed strings ("what are you changing?"), literal hole-question examples with a wrong/right contrast ("why would black cause that?" is safe; "why would black increase clicks?" is wrong), an exact allowlisted tool list, copy-paste-ready JSON contracts for both structured output and the handoff, and a fully worded nudge table. Every common case (junk, A/A, rewrite, scaffold, multiple measurements, injection input) has concrete handling. Not 4: there are no gaps — even fallback behavior ("If that skill isn't available, give them the handoff payload anyway") is specified.

5 / 5

Workflow Clarity

The process is clearly sequenced with explicit checkpoints and feedback loops: Step 0 routes in a fixed checked order with the reason for the order given; the conversational loop is grade → show sentence → one nudge → "let them fill the holes in their own words and grade again"; structured mode mandates lookups before the single JSON reply; and there are explicit stop conditions ("stop once it's strong", "Handoff, then stop", "Don't re-grade anything they didn't send") plus error-recovery paths (injection handling, missing-skill fallback). No destructive or batch operations exist, so the validation cap does not apply; this matches the top anchor's checklist and feedback-loop criteria.

5 / 5

Progressive Disclosure

The file is well-sectioned (route first, scaffold building, hole rules, fixed strings, modes, handoff, anti-patterns) with all navigation via clear headers, and every operational rule genuinely needs to be resident at runtime, so inline placement of the core is appropriate. Not 5: there are no bundle files at all while the skill runs ~190 lines, and content that could live elsewhere — the research backstory, provenance/sync comment, and changelog — is inlined, which is a minor organization gap rather than an ideal split.

4 / 5

Total

17

/

20

Passed

Description

75%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.

A strong, specific, third-person description that comprehensively covers the skill's actions, output modes, and safety boundary, and is clearly distinct from the sibling setup skill. Its main weakness is the absence of an explicit 'Use when...' trigger clause, which caps completeness and leaves a few natural trigger phrases (e.g. 'A/B test') unstated.

Suggestions

Add an explicit trigger clause, e.g. "Use when a user wants to write, improve, or grade an experiment hypothesis (A/B or A/A test) before setting it up in LaunchDarkly."

Include natural synonyms users would say, such as "A/B test" and "experiment idea", alongside "experiment hypothesis".

State the when for grading vs. building explicitly (e.g. "Use when someone brings a rough experiment idea or a draft hypothesis to evaluate") so the skill triggers on evaluation requests, not just authoring ones.

DimensionReasoningScore

Specificity

The description lists multiple concrete actions with comprehensive coverage: "Scores the draft against the Change / Measurement / Rationale rubric", "scaffolds an If/then/because sentence with fill-in questions", "detects junk input, A/A tests, and multiple measurements", "returns the experiment builder's JSON contract", "check existing flags and metrics with read-only lookups", and "hands off to launchdarkly-experiment-setup". It also states the boundary ("Never writes to LaunchDarkly"). Not 4: coverage spans every function of the skill including both output modes and the safety boundary, matching the 'comprehensive coverage' anchor rather than 'minor gaps'.

5 / 5

Completeness

The 'what' is clearly and fully answered (turn an idea into a testable hypothesis or grade a draft, with the rubric, scaffold, and detection rules), but there is no 'Use when...' clause or equivalent explicit trigger guidance — "when the caller asks for JSON only" is a mode-selection condition, not a usage trigger, so per the judging guideline completeness is capped at 3. This matches the anchor 'Has a clear what but when is missing or only weakly implied' (use is only implied by the subject matter).

3 / 5

Trigger Term Quality

Good natural-keyword coverage: users would say "experiment hypothesis", "grade one they already wrote", "A/A tests", "flags and metrics", "rough idea". Not 5: common synonyms a user would naturally say are missing — most notably "A/B test" / "experiment idea", and there is no phrasing like "when writing a hypothesis" as a raw trigger phrase. Not 3: the terms present go well beyond 'some relevant keywords' and include the domain's natural vocabulary.

4 / 5

Distinctiveness Conflict Risk

It occupies a clear niche (LaunchDarkly experiment hypotheses) with distinct triggers ("hypothesis", "A/A tests", "Change / Measurement / Rationale rubric") and explicitly differentiates from the nearest neighbor: "hands off to launchdarkly-experiment-setup" and "Never writes to LaunchDarkly". Not 4: the handoff and read-only boundary remove even the minor overlap risk with the setup skill.

5 / 5

Total

17

/

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/experiments/launchdarkly-experiment-hypothesis-builder/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.