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.
This is a strong, deeply actionable skill body: fully executable commands, mandatory build sequences, and unusually rigorous validation gates with feedback loops. The main weakness is token efficiency — the same engine-caveat material (NPV cache-0, iterate/write-order surgery) is repeated across three sections and version pins are scattered inline rather than consolidated.
Suggestions
Consolidate the NPV/XNPV cache-0 guidance into one canonical section (e.g., Known Issues) and have Recipe B Step 4 and 'Financial function patterns' reference it with a single line instead of re-explaining the SUMPRODUCT fallback three times.
Merge the write-order/iterate material in Recipe C Step 3 with the 'Write-order surgery' playbook under 'Circular references & iterative calc', keeping only a pointer in the recipe.
Move version-pinned caveats ('verified on v1.0.63', 'on v1.0.63+') into a dedicated version-compatibility note in Known Issues instead of repeating the version number inline, so stale pins are easy to update.
Consider splitting the three recipes into references/ files (three-statement.md, dcf.md, lbo.md) with a short in-body overview, since the 560-line single file is at the upper limit of navigable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly operational with no finance-theory padding ('runnable skeleton, not finance theory'), but guidance is systematically duplicated across sections — the NPV-caches-0 fix appears in Recipe B Step 4, twice in 'Financial function patterns', and again in 'Known Issues'; write-order/iterate surgery appears in both Recipe C and 'Circular references' — and the version pin 'v1.0.63' is cited five times inline with no deprecated/old-patterns section. This matches 'mostly efficient but could be tightened' rather than the minor-instances anchor. | 3 / 5 |
Actionability | Nearly every instruction is a copy-paste-ready command: complete heredoc `officecli batch` JSON with exact props and number formats, precise `--type`/`--prop` flags for named ranges, comments, validation dropdowns, and per-recipe verification one-liners. Specific examples cover the three common model cases end-to-end. | 5 / 5 |
Workflow Clarity | Each recipe has a mandatory build order with dependency rationale ('Do NOT build BS before P&L — RetainedEarnings depends on NI'), and the Audit & Delivery Gate section provides six executable gates with REJECT/exit-1 checks, error-recovery feedback loops (cache-refresh pass, de-ring/write/re-ring surgery, convergence verification), and an explicit 'assume there are problems' posture. | 5 / 5 |
Progressive Disclosure | Structure is good: base xlsx rules are delegated one level deep to skills/officecli-xlsx/SKILL.md via clearly signaled '→ see xlsx v2 §X' pointers, and the body keeps only financial-model deltas. However, no bundle files exist, so all three large recipes plus sensitivity/scenario and circularity playbooks live inline in a single 560-line file where a references/ split (e.g., per-recipe files) would ease navigation — a minor organization gap rather than misplaced content. | 4 / 5 |
Total | 17 / 20 Passed |