Content
80%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A dense, actionable skill body with excellent copy-paste recipes and no padding, covering the classification surface thoroughly. It loses points only on the multi-step server-log ingestion workflow lacking an explicit checkpoint sequence and on keeping large reference tables inline rather than splitting them into a one-level-deep reference file.
Suggestions
In 'Seeing bots that don't run JS', replace the implicit path with an explicit numbered sequence with a checkpoint, e.g. 1) confirm whether $http_log events are ingested, 2) if not, point the user to server-side capture (Vercel logs / edge worker / capture API), 3) only then build bot insights.
Move the virtual-property and HogQL-function reference tables into a single references file (e.g. references/classification-api.md) and link to it from SKILL.md so the main file is a lean overview with one-level-deep navigation.
Tighten the duplicated keep-vs-drop framing: keep the per-bucket 'default move' only in the traffic-types table and trim the overlapping bullets from the intro.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body assumes Claude's knowledge (no explaining what PostHog, HogQL, or analytics are) and goes straight to classification surfaces, tables, and copy-paste recipes; the only minor redundancy is the keep-vs-drop framing appearing in both the intro bullets and the traffic-types table, but it stays above the 'could be tightened' bar of a 2. | 3 / 3 |
Actionability | Provides fully executable, copy-paste-ready JSON property filters, TrendsQuery objects, and SQL with a real coalesce/nullIf user-agent expression, so it clears the 'fully executable, copy-paste ready' anchor rather than the pseudocode level of a 2. | 3 / 3 |
Workflow Clarity | Recipes are clear and unambiguous single actions (the simple-skill guideline permits a 3 for those), but the genuine multi-step path in 'Seeing bots that don't run JS' (no JS fired -> need server log ingestion -> then build bot insights) has an implicit rather than explicit ordered sequence with no validation checkpoint, landing at the 'steps present but checkpoints implicit' level of a 2. | 2 / 3 |
Progressive Disclosure | The file is well-organized with clear sections, tables, and recipes, but at ~177 lines it exceeds the under-50-line simple-skill carve-out, keeps the substantial property/function reference tables inline, and references only a single internal source path rather than signaled one-level-deep reference files, so it sits at 'some structure but content that should be separate is inline' rather than a 3. | 2 / 3 |
Total | 10 / 12 Passed |