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, unusually concrete orchestration skill: clear sequenced workflow, numeric thresholds throughout, and excellent reference signaling via the Reference Map and per-phase Read columns. The main costs are token waste from a duplicated HTML header comment, repeated TTL/routing/boundary content, and citation padding, plus missing worked examples for its two most common outputs (a METAPATTERNS.md entry and a LORE_INSIGHT message).
Suggestions
Delete or collapse the ~30-line HTML CAPABILITIES_SUMMARY comment block — it restates the body (capabilities, collaboration patterns, partners) at full length and is pure token overhead on every load.
State the Domain TTL multiplier table once (in Knowledge Decay Detection) and reference it from the Core Contract, and consolidate the routing/boundary guidance repeated across Trigger Guidance, Output Routing, and Collaboration into one section.
Add one worked METAPATTERNS.md catalog entry and one sample LORE_INSIGHT message so the most common outputs have copy-adaptable templates in the body.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly dense, project-specific protocol Claude would not already know, but there is real, repeated waste: the ~30-line HTML CAPABILITIES_SUMMARY comment duplicates body content wholesale, the Domain TTL multiplier table appears twice (Core Contract and Knowledge Decay Detection), and routing/boundary guidance is repeated across Trigger Guidance, Output Routing, and Collaboration. Justification prose ("Ebbinghaus curve", "MIT Sloan", "PMC meta-analysis", "temporal knowledge graph research shows") adds tokens without operational value. This fits 'mostly efficient but includes some unnecessary explanation or could be tightened' rather than 2, since the excess is duplication and citation padding rather than explaining concepts Claude already knows. | 3 / 5 |
Actionability | Guidance is highly concrete and executable for an instruction-only skill: exact clustering thresholds (>=80% / 50-79% / <50%), numeric freshness scoring with alert thresholds (<85%, <70%), confidence bands by evidence count (1=Anecdote ... 11+=Foundational), TTL multipliers per domain, explicit ID and journal formats ([DOMAIN]-[TYPE]-[NNN], '## YYYY-MM-DD - [Discovery/Insight]'), and a state-to-action table. It is not a 5 because there is no worked example of a METAPATTERNS.md catalog entry or a LORE_INSIGHT/LORE_ALERT message — the most common outputs have no template or sample inline. | 4 / 5 |
Workflow Clarity | The HARVEST → SYNTHESIZE → CATALOG → PROPAGATE → AUDIT workflow is clearly sequenced with per-phase required actions, key rules, and read references; destructive operations (archiving/removing patterns) are guarded by explicit validation gates (Ask First for <3 evidence, Never auto-archive FAILURE/ANTI by time alone, FOUNDATIONAL requires human decision) and the freshness table maps each state to a default action. It is not a 5 because error-recovery loops are thin — 'contradiction resolution workflow' is invoked by name with no inline steps, and audit findings route to a reference rather than a closed feedback loop. | 4 / 5 |
Progressive Disclosure | Structure and signaling are strong: a Reference Map with 'Read this when' guidance, per-recipe 'Read First' columns, per-phase 'Read' pointers, and references that appear one level deep. It is not a 5 because the referenced files (reference/*.md, _common/*.md) are not present in the bundle, so the navigation layer cannot deliver, and substantial detail is inlined in SKILL.md that clearly belongs in those references (the full Architecture node/edge enumeration, the TTL multiplier table duplicated inline despite reference/decay-detection.md being declared authoritative). | 4 / 5 |
Total | 15 / 20 Passed |