The body is lean and assumes Claude's competence, reserving prose for genuinely non-obvious PostHog specifics (the reserved 'web' type, the two distinct rollout_percentage fields) that Claude would not already know. It is not level 2 because there is no generic padding or restating of concepts Claude knows.

Provides a complete, copy-paste-ready experiment-create JSON block with concrete field names and explicit defaults ('50/50 control/test, 100% rollout'), matching the 'fully executable, copy-paste ready' anchor.

The 3-step flow is clearly sequenced with explicit delegation checkpoints ('Before asking any rollout question, load configuring-experiment-rollout') and a post-creation checklist; negative guidance ('Do NOT configure metrics on creation') acts as a validation checkpoint. Not level 2 because validation/sequencing is explicit rather than implicit.

With no bundle files present, the skill keeps its ~100-line overview inline and delegates detail to one-level-deep, clearly-signaled sibling skills via 'load the ... skill', with no nested references and well-organized sections.

Lists multiple concrete actions (defining the hypothesis, configuring rollout, setting up analytics) and explicitly names the two sub-skills it delegates to, matching the 'lists multiple specific concrete actions' anchor.

Explicitly states what the skill does (3-step creation flow) and when to use it via a 'TRIGGER when' clause plus a 'DO NOT TRIGGER when' exclusion, clearly answering both what and when.

Includes natural phrasings users would say ('create a new experiment', 'A/B test') alongside the tool name 'experiment-create', giving good coverage of natural terms.

The explicit negative triggers (updating existing experiments, managing lifecycle, browsing) and named delegation to sibling skills carve out a clear niche, making conflict with adjacent skills unlikely.