Content
71%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.
A well-structured, actionable body with concrete API calls, field shapes, and a clearly sequenced metric-building workflow guarded by pre-flight validation. Its main defect is that the three `references/*.md` files it points to for full schemas, templates, and result interpretation are absent, breaking the progressive-disclosure path.
Suggestions
Add the missing `references/` bundle files (`metric-templates.md`, `metric-configuration.md`, `interpreting-results.md`) the body already cites, or inline the critical bits and remove the dangling references.
Include at least one complete copy-paste-ready inline `ExperimentMetric` JSON payload (e.g. a ratio metric) so the skill is actionable even before the reference file is consulted.
Add an explicit post-update verification step (e.g. re-call `experiment-get` to confirm the metric attached with the intended type) to close the destructive replace-list workflow with a feedback loop.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient and product-specific (assumes Claude knows what APIs/flags are), but the WRONG/RIGHT example block and the extended bias-risk prose could be trimmed without losing the operational point. | 4 / 5 |
Actionability | Highly concrete — names exact endpoints (`experiment-saved-metrics-list?event=`, `experiment-update`), field shapes, and parameter names (`saved_metrics_ids`, `math: "sum"`, `allow_unknown_events`), but defers full inline JSON payloads to a reference file rather than giving copy-paste-ready payloads in the body. | 4 / 5 |
Workflow Clarity | Clear Step 1→4 sequence with pre-flight validation checkpoints ('always get the current experiment first via experiment-get', 'MUST call read-data-schema', 'confirm the match with the user') for the destructive list-replacing operations; minor gap is the absence of an explicit post-update verification step. | 4 / 5 |
Progressive Disclosure | Structure and signaling are good — clear sections and one-level-deep references to `references/metric-templates.md`, `references/metric-configuration.md`, and `references/interpreting-results.md` — but the `references/` directory does not exist, so those signaled references dangle and the disclosure is non-functional. | 3 / 5 |
Total | 15 / 20 Passed |