Content
82%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 tightly engineered operational core: every section is executable configuration rather than explanation, with an unusually strong safety spine (tiered risk model, tool lockout, verbatim scripts) and a genuine validate-and-recover loop. The main residual issues are minor redundancy between the precedence ladder and the risk model, and two bundle files missing from the root load order.
Suggestions
Consolidate the tool-lockout/'code waits' language between Section 2 (Precedence Ladder) and Section 5 (Tier 1/Tier 2 rows) so each rule is stated once and cross-referenced.
Add `references/model-calibration.md` and `references/crisis-sources.json` to the Section 1 load-order list (e.g., under a 'Calibration & Verification' entry) so every bundle file is discoverable one level deep from SKILL.md.
Either inline one sentence per step of the 6-step turn sequence (what 'Formulate' or 'Anchor' concretely requires) or explicitly state that clinical-spine.md is mandatory reading before the first full session.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and assumes Claude's competence — no space is spent explaining what grief, IFS, or polyvagal theory are — and safety-critical data like the helpline table earns its tokens. Minor duplication keeps it below anchor 5: "Lock all tools"/"code waits" in the Precedence Ladder repeats the Tier 1/Tier 2 rows, and "stay present" appears twice within the Tier 2 bullet. | 4 / 5 |
Actionability | Guidance is copy-paste ready for an instruction-only skill: the exact safety question ("When you say that, are you thinking about ending your life?"), verified helpline numbers per country, a fully specified SATORI_NOTE format with per-field char limits, the verbatim trust warning, and concrete anti-slop examples ("clarity, purpose, and peace", "Here's the thing"). The precedence ladder enumerates every input class with an unambiguous behavior. | 5 / 5 |
Workflow Clarity | The precedence ladder is a clear ranked decision sequence re-evaluated "every turn", and the honesty self-check plus Recovery Rule ("If any check fails, revise before sending") forms an explicit validate-then-fix feedback loop. Minor validation gaps hold it at 4: the six-step turn sequence ("Attune → Clarify → Formulate → Integrate → Translate → Anchor") is only named, with execution checkpoints delegated to clinical-spine.md, and the body never links the helpline verification source. | 4 / 5 |
Progressive Disclosure | A canonical sub-8KB overview with a numbered, category-labeled load order linking 16 reference files — all of which exist and are one level deep. Two minor organization gaps: `crisis-sources.json` and `model-calibration.md` are absent from the load order and only discoverable via clinical-spine.md, and reference files cross-link each other (e.g., traditions-quickref.md → traditions-eastern.md), adding mild navigation depth. | 4 / 5 |
Total | 17 / 20 Passed |