Content
57%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is a strong, highly executable reference with real copy-paste code for every major PostHog surface, and it stays focused on project-specific knowledge rather than generic tutorials. Its weaknesses are structural: it is a monolithic 950-line file with no reference files, the batch dashboard-creation workflow lacks any validation checkpoints, and there is enough duplication and teaching-of-known-conventions to warrant trimming.
Suggestions
Move the per-vertical dashboard checklists (~200 lines) and the core-events catalog into reference files (e.g. references/dashboards.md, references/events.md) linked one level deep from SKILL.md — the all-inlined monolith is what holds progressive_disclosure at 3.
Add validation checkpoints to the MCP dashboard workflow (e.g. verify the dashboard was created and the insight query returns rows before adding it to the dashboard), since batch-creating dashboards and insights without verification caps workflow_clarity at 3.
Cut the good/bad event-naming examples and the duplicated dashboard summaries (the Quick Reference table repeats the Project-Specific Dashboards section), and fix the Python snippet's missing 'import os' to tighten conciseness and close the actionability gap.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Most of the body is concrete, project-specific code rather than padded explanation, but it could be noticeably tightened: the event-naming good/bad lists ('Too vague', 'Not snake_case', 'Not past tense') teach conventions Claude already knows, and dashboard guidance appears three times (per-vertical checklists, the MCP workflow, and the summary table). Between anchors 2 and 3 — the bulk is genuinely useful material, so it clears 'noticeably verbose' but not 'efficient'. | 3 / 5 |
Actionability | Nearly all guidance is copy-paste-ready executable code covering install, init, identification, tracking, flags, sanitization, and E2E mocking across Next.js, React, Python, and Node. It stops short of anchor 5 due to minor gaps: the Python snippet uses os.environ without importing os, and the mcp_posthog_* calls in the dashboard workflow are illustrative pseudo-invocations rather than exact commands. | 4 / 5 |
Workflow Clarity | The MCP dashboard workflow has a real numbered sequence ('First, check existing dashboards' through '4. Set up key funnels'), but no validation or verification checkpoints anywhere — nothing confirms a dashboard or insight was created correctly before proceeding. Since the workflow batch-creates multiple dashboards and insights, the rubric's cap for batch operations without validation applies, holding this at 3 rather than 4. | 3 / 5 |
Progressive Disclosure | No bundle files exist (no references/, scripts/, or assets/), so everything is inlined in a single ~950-line SKILL.md. Section headers do provide navigable structure (better than anchor 2's 'minimal structure'), but roughly 200 lines of per-vertical dashboard templates and the full event catalog are content that clearly belongs in separate reference files — exactly anchor 3's 'content that should be separate is inline'. | 3 / 5 |
Total | 13 / 20 Passed |