Content
71%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 an unusually well-crafted operational core: precise crisis protocols with verified helplines, a clear precedence ladder, an executable continuity-note format, and a built-in validation loop. Its main weakness is that the entire progressive-disclosure promise is unbacked — none of the referenced files exist in the bundle, so the deferred core content (clinical spine, protocols, toolkits) is unreachable, and the Turn-1 self-sufficiency claim does not hold for anything beyond the immediate crisis rules.
Suggestions
Ship the references/ bundle: all ~16 files cited in §1 (SOUL.md, clinical-spine.md, the four traditions files, both toolkits, and the seven protocol files) are missing, so every deep link in the body is broken.
Add one-line definitions of each step of the 6-step sequence (Attune → Clarify → Formulate → Integrate → Translate → Anchor) directly in the body so sessions work before any reference loads, making the Turn-1 self-sufficiency claim true.
De-duplicate the tier behavior descriptions between the §2 Precedence Ladder and the §5 Three-Tier Risk Model — state each tier's behavior once and cross-reference it from the ladder to trim tokens.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and functional with no padding of concepts Claude already knows; the region-specific helpline table is exactly the kind of hallucination-prone data that earns its tokens. It is not 5 because Tier 0/1/2 behaviors are stated twice (the §2 Precedence Ladder restates §5's tier definitions), and atmospheric phrasing ("warm companion thinking with clinical rigor and speaking with human ease") adds a little redundancy. | 4 / 5 |
Actionability | Highly concrete in the right places: the exact direct safety question ("When you say that, are you thinking about ending your life?"), verified per-country helpline numbers, a copy-paste SATORI_NOTE v6.5 template with field limits, and a 4-question pre-send honesty check with a recovery rule. It is not 5 because the core 6-step turn sequence ("Attune → Clarify → Formulate → Integrate → Translate → Anchor") names its steps but defers how to execute each to clinical-spine.md, and some directives ("quiet spacious presence") remain abstract. | 4 / 5 |
Workflow Clarity | Sequences are clearly ranked and re-evaluated per turn ("re-evaluate every turn; escalate immediately if new risk emerges"), and §6 is a genuine validate→fix→resend feedback loop ("If any check fails, revise before sending"). It is not 5 because the per-step checkpoints for the central turn sequence live in a deferred reference rather than the body, leaving the session-level workflow's checkpoints implicit. | 4 / 5 |
Progressive Disclosure | The in-file design matches the top anchor: a numbered load order, well-signaled one-level-deep references grouped by purpose, and a compact-mode fallback ("Context-constrained models load only references/core-compact.md"). However, scoring against the actual bundle structure, the skill ships no references/ directory at all — every one of the ~16 referenced paths (SOUL.md, clinical-spine.md, all protocols and toolkits) is broken, so the claimed structure is unfulfilled and the body is not actually self-sufficient on Turn 1 as claimed. This lands between the broken-navigation penalty and the excellent in-file organization. | 3 / 5 |
Total | 15 / 20 Passed |