Content
93%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-engineered skill body: lean, concrete, and self-verifying through blocking exit codes and an enumerated gate checklist, with an appropriately split bundle and clear navigation. The only notable gap is the absence of an explicit fix-and-re-run feedback loop around the exit-4 gates.
Suggestions
Add one line after the gates section stating the recovery loop explicitly, e.g. 'On exit 4, fix the listed condition (assign an owner, book the entry, quantify the explanation) and re-run — exit 0 is the only pass.'
Align the workflow example filenames with the shipped example assets (use --plan assets/close_plan.example.json or note that close_plan.json follows assets/close_plan.example.json) so the commands run verbatim against the bundle.
State what exit code 2 ('at risk' / 'items' / 'evidence') means for the user — whether it requires action or is informational — since only exit 4's conditions are enumerated.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~80-line body is lean and every section is functional: three commands, terse step explanations, a gates checklist, an exit-code table, and an annotated file listing. It assumes Claude's competence — it states this skill's rules and script behavior rather than explaining what a reconciliation or critical path is. | 5 / 5 |
Actionability | All three scripts get fully executable commands with realistic argument values (--gl-balance 52708.60, --as-of 2026-10-31, --pct 10), an exit-code table per script, and a note that all support --sample and --output json, so the common cases are copy-paste runnable. | 5 / 5 |
Workflow Clarity | The three steps are clearly sequenced with one command each, and validation is explicit (numbered blocking gates plus per-script exit codes including '4 blocked'). It falls between anchors 4 and 5 because error recovery is a checklist of conditions rather than a feedback loop — the gates say what blocks but never say 'fix the item and re-run until exit 0'. | 4 / 5 |
Progressive Disclosure | SKILL.md is a concise overview with well-signaled, one-level-deep references: the 'References and assets' section annotates each of the three reference files with its scope and source count, and all eight referenced bundle files (references, scripts, assets) exist on disk with no nested reference chains. | 5 / 5 |
Total | 19 / 20 Passed |