Content
76%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-architected multi-phase skill with explicit sequencing, validation gates, loud-failure handling, and clean one-level-deep reference splitting. The main improvement is trimming the repetitive justifications of the question-format rules to recover token budget.
Suggestions
Collapse the three-sentence justification of the 'repeat the counts inside the question' rule (Step 5) into one directive sentence; the rule is already clear from its statement.
Dedupe the FR-UNIVERSAL.QA-FORMAT self-containment guidance, which is restated in both the Question Format section and again inside Step 6.
Add a one-line concrete example of the subagent spawn invocation (e.g. the Task/Agent tool call shape) so the fan-out step is copy-paste ready across IDEs.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient but several passages pad the same constraint repeatedly — the 'repeat the counts inside the question' rule is justified across three long sentences (Step 5) and the self-containment rule is restated in multiple places — so it could be tightened without losing clarity. | 3 / 5 |
Actionability | Provides concrete executable guidance — specific fan-out pattern (5 parallel maintenance-scan-* subagents across buckets W1–W5), exact issue-line shape, literal severity strings, sorting/numbering rules, and a worked example — with only minor gaps around how to actually spawn the subagents in each IDE. | 4 / 5 |
Workflow Clarity | A two-phase process (Scan, Resolution) is laid out as an explicitly numbered 6-step sequence with a mandatory verification gate (Step 3), a loud-failure 'NOT SCANNED' fallback, anti-inflation severity calibration, and a closing verification checklist with feedback loops for a destructive/batch operation. | 5 / 5 |
Progressive Disclosure | SKILL.md carries only the index and routing; the per-category check detail is pushed one level deep into the 5 self-contained subagent bodies, and severity/verification/example references all resolve to real files in references/ (severity-rubric.md, verification-gate.md, example-findings.md), each clearly signaled and linked. | 5 / 5 |
Total | 17 / 20 Passed |