Content
85%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 exceptionally actionable, well-sequenced workflow skill: concrete templates, exact tool names and paths, validation checkpoints, error-recovery loops, and a completion checklist. Its weaknesses are a monolithic single-file structure with no progressive disclosure into reference files, and noticeable repetition of the same critical rules across sections.
Suggestions
Split content that does not need to sit in the main workflow into reference files — e.g., references/file-templates.md (touch points 0–10 templates), references/mapping-heuristics.md (the role/signal table and confidence levels), references/identity-rules.md (the buildMemberIdentities case table) — and keep SKILL.md as a phased overview with one-level-deep pointers.
Consolidate repeated rules: state "Never pass lfUsername as platformUsername" once (touch point 9 or §3a, not both plus a third restatement), and let the Anti-Hallucination Rules section be the single home for "read before modifying" instead of repeating it in the File Inventory and Phase 4.
Replace the DOT digraph — which duplicates the phase headings that follow it — with a one-line phase sequence (Phase 1 → 5 with the two confirmation loops noted inline) to save ~25 lines.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Nearly every section carries repo-specific facts Claude cannot know (exact paths, MCP tool names, existing-source patterns), so the length is mostly earned. However, there is trimmable duplication: "Never pass lfUsername as platformUsername" appears three times (§3a intro, §3a critical note, touch point 9), "read before modifying" is stated three times, the DOT graph restates the phase headings, and "Show the full generated file and ask for confirmation" repeats per touch point. This is the 4 anchor (efficient with minor instances of over-explanation), not 5; not 3 because the bulk is non-derivable procedural content rather than explanation of concepts Claude already knows. | 4 / 5 |
Actionability | Fully executable throughout: exact MCP tool names (mcp__claude_ai_LFX_BI_Layer__get_source_details), runnable SQL templates for schema/sample/ambiguity resolution, TypeScript enum/grid/registration templates, exact shell commands (date +%s, pnpm format, pnpm lint), and exact repo paths for all 11 touch points. The {PLATFORM}-style placeholders are justified for a scaffolding skill. Matches the 5 anchor (copy-paste ready, covers the common cases). | 5 / 5 |
Workflow Clarity | Five explicitly sequenced phases with a flow diagram, decision branches (Case A/B, Option A/B), per-mapping confirmation checkpoints, an explicit error-recovery loop ("If any issue found: loop back to the relevant Phase 3 sub-section, fix the mapping, regenerate the affected file"), a lint-until-clean loop, and a final completion checklist. This matches the 5 anchor (clear sequence, explicit validation, feedback loops, checklists). | 5 / 5 |
Progressive Disclosure | No bundle files exist (references/, scripts/, assets/ are absent); all ~730 lines are inline in SKILL.md, including content that belongs in separate reference files (per-touch-point file templates, the role-heuristics table, migration SQL patterns). Section headers are clear and navigable — so not the 2 anchor's "minimal structure" — but content that should be separate is inline with no references at all, matching the 3 anchor rather than 4. | 3 / 5 |
Total | 17 / 20 Passed |