Content
66%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 SKILL.md body is an exemplarily structured thin index: clear mode detection, well-sequenced per-mode workflows with explicit gates and a Definition of Done, and lean one-liner principles with no padding. Its critical weakness is that the bundle ships only 2 of the 14 files it references — all rules/*.md and templates/* are missing — so the progressive-disclosure structure and the per-mode workflows dead-end on execution, leaving the detailed guidance (PII gate, implementation rules, wrapper template) undeliverable.
Suggestions
Ship the referenced bundle files: add the 10 rules/*.md files (what-to-track, pii-and-compliance, implementation-web/mobile, etc.) and both templates/* files (analytics-wrapper.template.ts, tracking-plan.template.yaml) — without them every mode's first step fails and progressive disclosure is non-functional.
Add one small inline executable example in implement mode (e.g., a minimal typed track() call through the centralized wrapper) so the skill retains actionable guidance even before the referenced files load.
De-duplicate the file lists: either drop the 'Required Reading by Mode' table or remove the per-mode file references from 'Workflow by Mode', keeping one canonical navigation path.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is an efficient thin index with no concept-explanation padding (no 'what is RUM' filler), dense one-liner principles and anti-patterns, and every section serving navigation. Minor trims possible: the "Required Reading by Mode" table duplicates the per-mode file lists already given in "Workflow by Mode", and the external-dependency callout ("and otherwise skips them with a one-line report, never silently") is wordy — matching anchor 4's 'minor instances that could be trimmed', not anchor 3's unnecessary explanation. | 4 / 5 |
Actionability | Process guidance is specific ("gate the user before writing instrumentation", "≤ 30 event names", named codegen tools "Avo, RudderTyper, Typewriter", "Mark deprecated first, set a sunset date"), but the body contains no executable code or commands and delegates all concrete payloads to files that are missing from the bundle — following "Load rules/what-to-track.md" (step 1 of guide mode) or the wrapper template dead-ends. This matches anchor 3: concrete guidance present but incomplete, key details missing; it exceeds anchor 2's 'high-level hints only' but falls short of anchor 4's 'concrete code or commands with minor gaps'. | 3 / 5 |
Workflow Clarity | Each of the five modes has a numbered sequence with explicit conditional branches and validation checkpoints: "gate the user before writing instrumentation", "Run the PII gate ... on every property before the diff is final", "Identify downstream consumers ... before deletion", "do not auto-edit unless the user approved a pre-defined audit scope", plus mode detection with a stated default and a Definition of Done checklist. The destructive remove mode is properly guarded (deprecate → check consumers → delete), so the destructive-operation cap does not apply; this matches anchor 5's 'explicit validation steps ... checklists for complex processes' rather than anchor 4, which reserves room for missing checkpoints that are not missing here. | 5 / 5 |
Progressive Disclosure | The index design itself is well-signaled and one level deep (SKILL.md → rules/references/templates, with a per-mode loading table), but scored against the actual bundle only 2 of the 14 referenced paths exist (references/event-catalog.md, references/platforms.md); all 10 rules/*.md files and both templates/* files are absent, so every mode's step-1 load fails and the structure is non-functional as shipped. This is a navigation failure of comparable severity to anchor 2 — 'references are buried'/minimal functional structure — rather than anchor 3's mere organizational improvement, and well below anchor 4's 'references mostly clear' because most references do not resolve at all. | 2 / 5 |
Total | 14 / 20 Passed |