Content
67%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-structured persona skill with genuinely actionable guidance: explicit classification cues, per-question-type response structures, a safety read-before-responding checkpoint, and an exemplary reference-file table whose paths all resolve. The main weaknesses are token padding — an abundance of voice exemplars, redundant evidence-tier instructions, and a speculative module tree — that could be tightened or offloaded to the reference files.
Suggestions
Trim the seven block-quote voice exemplars in 'The Mentor's Voice' to two or three, and move the extended register discussion (e.g., the historical-perspective pacing rules) into references/persona-guide.md, keeping only the core register definition in SKILL.md.
Consolidate the evidence-quality signaling: define the tier system once (either in Step 3 or in references/evidence-grading.md) and let the format rules reference it in one line instead of re-explaining how to express it in prose.
Remove the 'Module Expansion' tree lines for '(planned)' files (drug-interactions.md, specialties/) — advertising not-yet-existing bundle files adds tokens and misleads navigation — or move the tree to the hub-level documentation.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient persona guidance, but there is tightening to do: seven extended block-quote voice exemplars where two or three would calibrate the register; a "Module Expansion" tree listing "(planned)" files that is speculative and not actionable; a "Source" GitHub link line; and redundancy between Step 3's 🟢🟡🔴⚪ tier definitions and the later "No emoji-based evidence grading in running text" rule, which re-explains the same system. This fits anchor 3 ('mostly efficient but includes some unnecessary explanation or could be tightened') — clearly more substantive than anchor 2's padded sections, but not anchor 4's lean fit. | 3 / 5 |
Actionability | For an instruction-only skill, the guidance is largely concrete and executable: student-level classification with explicit lexical cues ("uses 'differential,' 'NNT,' 'GRADE', cites trials"), a per-question-type response structure block, named reference-file read triggers, a ❌/✅ paired metformin example showing the target register, a defined fallback ("default to intermediate and offer: 'Want me to go deeper into the mechanism...'"), and concrete evidence-tier definitions. It falls short of anchor 5 because some voice guidance is abstract and unverifiable ("let the prose breathe", "sentences can be longer, more contemplative") with no beginner-level worked example, matching anchor 4 ('mostly executable guidance... with minor gaps'). | 4 / 5 |
Workflow Clarity | The Response Workflow is a clear four-step sequence (classify → structure → evidence signaling → GRADE) with a decision table mapping student level and question type to behavior, and a genuine safety checkpoint ("For personal health context ('I have...', 'my child...') → Read references/safety-framework.md before responding"). Not a destructive/batch operation, so no validation cap applies. It misses anchor 5 because there is no error-recovery or verification loop (e.g., what to do when classification is ambiguous beyond the one fallback line), and Steps 3 and 4 partially overlap on evidence-quality handling — anchor 4's 'clear sequence with most checkpoints present; minor validation gaps'. | 4 / 5 |
Progressive Disclosure | The opening Reference Files table with a "When to read" column per file is exemplary well-signaled navigation, and all three referenced paths (evidence-grading.md, persona-guide.md, safety-framework.md) exist as one-level-deep bundle files — verified on disk. Kept below anchor 5 by two organization gaps: the ~100-line persona voice section is substantial content that could largely live in persona-guide.md (the SKILL.md keeps only the core), and the module tree advertises not-yet-existing "(planned)" files. Anchor 4 ('good structure; most content appropriately placed; references mostly clear; minor organization gaps') is the best fit. | 4 / 5 |
Total | 15 / 20 Passed |