Content
78%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 an exceptionally lean, well-sequenced audit checklist with appropriate safety constraints and correct read-only framing. Its main weakness is actionability: it tells Claude what to inspect but not how — no tool paths, no deduplication-check method, and no output schema or example despite requiring 'schema-valid findings'.
Suggestions
Add concrete execution pointers for the verification steps, e.g., where to check deduplication (match event_id between browser pixel and CAPI/server payloads in sGTM preview or Meta Events Manager Test Events) and where to find destination diagnostics.
Include the findings output schema or a one-line example finding object so 'Return schema-valid findings, failure modes, owner, priority, and verification steps' is unambiguous.
Show a brief example of the browser-vs-server comparison in step 2 (a short field-by-field list such as event_name, event_id, value, currency) so the comparison has a concrete artifact to check against.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is ~19 lines with zero padding: no explanation of what sGTM or conversion APIs are, no generic filler, and every line either scopes a step or adds a non-obvious constraint ("hashing does not make unnecessary data collection acceptable", "Never retain raw identifiers or credentials in artifacts"). It fully matches the 'lean and efficient; every token earns its place' anchor. | 5 / 5 |
Actionability | Steps name concrete objects of inspection ("event taxonomy, parameters, IDs, timestamps, values, currency, user data, and consent state") but give no executable specifics: no tools or locations (sGTM preview, Meta Events Manager Test Events), no method for verifying deduplication (e.g., matching event_id between pixel and CAPI payloads), and no schema or example for the required "schema-valid findings". This sits above anchor 2 (pure high-level hints) but squarely in anchor 3: some concrete guidance with key execution details missing. | 3 / 5 |
Workflow Clarity | The six steps form a coherent sequence (map → compare → verify → inspect → reconcile → return), and step 5 ("Reconcile destination diagnostics with source logs and business conversions") plus step 6's required "verification steps" act as checkpoints. It falls short of anchor 5 because there are no explicit feedback loops or error-recovery guidance, but it is above anchor 3 since checkpoints are named rather than absent; the read-only constraint ("Do not change production tracking") appropriately de-risks the operation. | 4 / 5 |
Progressive Disclosure | Per the rubric's guideline for skills under 50 lines with no need for external references, the body qualifies for the top score: it is a single well-organized unit (title, numbered process, closing untrusted-data/safety note) with nothing that belongs in a separate file and no nested references. | 5 / 5 |
Total | 17 / 20 Passed |