Content
81%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.
An operationally strong playbook: exact MCP call patterns, formulas, per-duration due/rate rules, and a sequenced workflow with real validation checkpoints and guardrails against known wrong methods. The main improvement opportunities are consolidating the repeated anti-pattern warnings and providing one complete runnable script template.
Suggestions
Consolidate the blended-churn / second-model warning, which is restated in the churn line, short horizons, steps, and wrong-example sections, into a single authoritative guardrail statement.
Provide one complete executable script template in the Correct block that wires the Status, Retention/Cohort, and MRR Movement pulls into the step loop, instead of schematic fragments.
Show the inflow run-rate derivation (the MRR Movement pull and how new/resub/expansion become inflow[m]) with the same concrete get-chart-data example given for Subscription Status.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and directive with no explanation of concepts Claude already knows, but the anti-pattern warning about blended churn / second-method projections ("never as a second method that sets the headline number") is repeated across at least four sections and could be consolidated. | 4 / 5 |
Actionability | Highly concrete guidance — a full get-chart-data call with exact parameters (segment=expiration_month, selectors, limit_num_segments), the churn formula, and a per-duration due/rate table — but the Correct script blocks are schematic (e.g., "due_annual_next = 28316.52", "inflow_run_rate") rather than a copy-paste runnable template wired to the chart pulls. | 4 / 5 |
Workflow Clarity | The MRR Steps are clearly sequenced (pull stocks, pull due, pull rates, pull inflows, compute in one script, present) with explicit validation checkpoints ("If month-1 modeled churn ≠ observed, print the gap as a diagnostic only", "Every rate literal in code must appear in a prior tool output") and an error-recovery path ("If you cannot pull due at all, ask for the renewal schedule"), plus a routing section that dispatches each ask type to the right workflow. | 5 / 5 |
Progressive Disclosure | A clear routing overview ("Pick the section that matches the ask") and well-organized sections, with all content appropriately inline for a single-task skill with no bundle files; however at ~172 lines with no reference layering, the wrong-example gallery and duration table could partially live in a reference file. | 4 / 5 |
Total | 17 / 20 Passed |