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.
A well-engineered instruction skill: a clearly gated five-phase workflow with real validation checkpoints, concrete output schemas and defaults, and excellent reference-file structure — all five bundle references exist and are one level deep. The one meaningful weakness is redundancy: three pairs of sections (tool tables, inputs vs. task questions, quality checks vs. common mistakes) restate the same guidance and could be merged to cut roughly 60–80 lines without losing any information.
Suggestions
Merge "Tool Selection Quick Picks" and "Tool Integrations" into one table (add MCP/guide-link columns to the quick-pick rows); both tables currently describe ~12 of the same tools twice.
Delete the "Task-Specific Questions" section — it restates "Inputs to Collect" nearly item-for-item; keep the single "ask once, then infer defaults" list instead.
Consolidate "Quality Checks" and "Common Mistakes": fold the non-overlapping mistakes (e.g., mixing branches, ignoring quiet hours) into the checklist so each rule appears once.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and free of concept-explainer padding (no "what is GDPR" or library tutorials), but three section pairs carry substantially duplicated guidance: "Tool Selection Quick Picks" vs "Tool Integrations" (two tables covering ~12 of the same tools), "Inputs to Collect" vs "Task-Specific Questions" (near-verbatim rephrasing of the same seven items), and "Quality Checks" vs "Common Mistakes" (~half the items overlap). This fits anchor 3 ("mostly efficient but... could be tightened") rather than 4, since the redundancy spans whole sections, not minor instances; it is well above 2 because every line still carries skill-specific, non-generic guidance. | 3 / 5 |
Actionability | For an instruction-only skill the guidance is fully executable: exact scoring rubric rows (Hot/Warm/Cold/Skip with definitions), literal CSV column schemas per branch, markdown table headers, concrete defaults ("default 25 for SaaS / B2B, 15 for Local SMB", "Source 2–3× more candidates", "~20% Hot, ~30% Warm"), and a per-tool task map. Specific examples cover the common cases (all three branches plus the no-enrichment-tools fallback), matching the anchor-5 coverage pattern. | 5 / 5 |
Workflow Clarity | Five explicitly sequenced phases with real validation gates for a batch operation: "Don't move to discovery without this" (ICP gate), "never assert without backing" evidence rules, "always verify deliverability before adding to the final list", an 8-item pre-finalization Quality Checks checklist, and error-recovery routing (failed-verification emails moved to a separate "invalid" bucket and flagged, unverified items surfaced under "Open questions"). This matches the anchor-5 pattern of explicit validation steps, feedback loops, and checklists; the batch-operation cap of 3 does not apply since validation is present at every stage. | 5 / 5 |
Progressive Disclosure | The SKILL.md is a genuine overview (shared framework, quick picks, formats) with well-signaled one-level-deep references: all five referenced files (saas/b2b/local-prospecting.md, data-sources.md, compliance.md) exist in the bundle, contain real content, and are linked at the point of need with descriptive framing. No nested references and no wall-of-text inline detail, matching the anchor-5 structure. The ../../tools/REGISTRY.md and integrations links sit outside this bundle and could not be verified here, but they are an external suite-level registry, clearly labeled, not a second level of skill detail. | 5 / 5 |
Total | 18 / 20 Passed |