Content
75%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.
The body is a tightly written, highly actionable control file: concrete phrasings, exact safety data, validation loops, and a clear precedence model. Its critical flaw is structural — the entire reference bundle it orchestrates does not exist, so the progressive-disclosure architecture and the detailed workflows it points to are unreachable.
Suggestions
Ship the references/ bundle that Section 1 promises (SOUL.md, clinical-spine.md, the tradition/toolkit files, all protocols, onboarding, tone-and-voice, memory-system, and core-compact.md), or inline the essential content into SKILL.md so the skill functions standalone.
Define the six-step turn sequence (Attune → Clarify → Formulate → Integrate → Translate → Anchor) in the body or in an existing clinical-spine.md file — currently only the names appear, so the core turn workflow cannot be executed from the canonical file.
Consider moving the per-country helpline table into a crisis reference file (keeping only the load trigger and one universal fallback like findahelpline.com inline) to reduce canonical-file bulk — but only once the reference bundle actually exists, since crisis-time data must remain immediately reachable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Dense, operational sections — precedence ladder table, tier model, note template, anti-slop prohibitions — with no explanations of concepts Claude already knows. Not 5: the ten-country helpline table is bulk data that could largely live in a crisis reference file, and a few meta lines ("Turn 1 is self-sufficient from this `<8 KB` canonical file") could be trimmed. | 4 / 5 |
Actionability | For an instruction-only skill the guidance is fully executable: a verbatim safety question ("When you say that, are you thinking about ending your life?"), exact helpline numbers per country, a copy-paste-ready SATORI_NOTE template with field length constraints, named prohibited phrases, and explicit condition-to-behavior mappings. Specific examples cover the common cases. | 5 / 5 |
Workflow Clarity | A clear sequence is present: sense risk tier, apply the precedence ladder ("re-evaluate every turn; escalate immediately if new risk emerges"), run the turn sequence, then the 4-question honesty self-check with an explicit recovery loop ("If any check fails, revise before sending"). Not 5 because the six-step Attune→Anchor sequence is only named in the body — its execution detail is deferred to a reference file. | 4 / 5 |
Progressive Disclosure | Section 1 lays out a well-signaled, one-level-deep load order with markdown links, but every one of the ~17 referenced files is absent — no references/ directory exists in the bundle, so SOUL.md, clinical-spine.md, all protocols, toolkits, traditions, and the compact fallback are dangling links. The hub-and-spoke design is good on paper, but with the spokes missing the disclosure structure collapses and the skill is unnavigable beyond the inline summary. | 2 / 5 |
Total | 15 / 20 Passed |