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.
A well-built operational skill: deterministic scoring tools, a defined workflow with routing, and exemplary progressive disclosure with a clean one-level-deep bundle. The main cost is redundancy — the voice rules are enforced in three separate sections, which inflates token cost without adding new guidance — and the workflow lacks an explicit recovery path for ambiguous inputs or disputed scores.
Suggestions
Merge 'Hard Rules' 3–7 and the overlapping 'Anti-Patterns' entries into the Operating Prompt section (or cut the restatements down to a single cross-reference) — the voice rules currently appear three times, which is the body's largest avoidable token cost.
Trim the positioning prose in the intro ('Where a generic advisor balances and reassures...') and the 'Andreessen-lens counterpart' paragraph to one sentence; the portability blockquote plus the workflow already establish what the skill is.
Add a short error-recovery step to the workflow: what to do when tool inputs are ambiguous or unavailable (e.g., 'ask the user for the missing score, default to unknown, or derive it from their answers') and how to handle a user disputing a scored verdict.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient — the workflow, forcing questions, and tool sections are tight — but the behavioral rules are restated three times (the Operating Prompt binding rules, Hard Rules 3–7, and Anti-Patterns), and the positioning intro ('Where a generic advisor balances and reassures, this skill takes a position') is padding that could be trimmed. Not a 2 because each section still carries operational content rather than concept explanation Claude already knows. | 3 / 5 |
Actionability | Fully executable guidance: copy-paste-ready invocations for all three scripts with flags and expected outputs, a routing table mapping user intent to tools, six forcing questions each with a recommended answer, and a defined verdict vocabulary (BUILD-POUR-FUEL / MARKET-FIRST-DERISK / KILL-OR-REPICK-MARKET). The common cases are covered. | 5 / 5 |
Workflow Clarity | A clear four-step sequence (route → interrogate one question at a time → score with tools → deliver verdict) with deterministic checkpoints built into the scripts (sub-4 market hard kill gate, Sean Ellis 40% gate). Not a 5 because there is no explicit error-recovery loop — e.g., what to do when inputs are ambiguous or the user disputes the scored verdict beyond 'don't capitulate without new evidence'. | 4 / 5 |
Progressive Disclosure | The body is a genuine overview: every referenced file exists (verified against the bundle listing), references are one level deep with no chained indirection, and each link is annotated with its purpose and source count. Verbatim prompt, canon, worksheet, and worked examples are correctly split into references/ and assets/ rather than inlined. | 5 / 5 |
Total | 17 / 20 Passed |