Content
23%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 rigorous, internally consistent formal contract whose workflow is more precisely specified than most skills, but it is delivered at extreme token cost and in a register that buries the executable procedure. A reader gets the contract's guarantees yet almost no concrete, copy-followable guidance for producing an actual round, and nearly all of it sits inline in SKILL.md instead of in the references layer.
Suggestions
Move the Lean formal block (FLOW, MORPHISM, the namespace development) into references/ (e.g. references/contract.lean) and keep in SKILL.md a plain-language statement of the protocol's core loop and the occasions on which the contract should be loaded, cutting the always-loaded body by an order of magnitude.
Add a worked example — one full round showing the intent relay, two or three concrete versions, the standing, and the opening — so the abstract contract is anchored to something executable.
Restate the core sequence as a short numbered procedure in everyday language (check activation → relay intent → produce → present round → read the turn → carry out asks → branch) before any formal vocabulary appears, so the workflow does not have to be mined from the Lean development.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is ~43KB (roughly 11k tokens) loaded on every activation, dominated by a Lean 4 formal apparatus (Turn/Cite/Coord/Occ, structural recursion, classical choice) that re-encodes ordinary conversational state management, with single doc comments running past 100 words (e.g. the notActivated, Reaches, and ClosingSupported axioms). Though the material is original rather than explanation of known concepts, the magnitude and density of the padding place this at anchor 1 ('severely verbose… heavily padded') rather than 2. | 1 / 5 |
Actionability | There are no commands, tools, or worked examples of what a relay, version, or round looks like in practice; guidance like "Compose each round so the reader can act without reassembly" and "realize each in a medium in which the qualities the person's judgment concerns can be encountered" stays at the high-hint level. The TOOL GROUNDING entries (e.g. the intentRelay and present strings) are semi-operational checklists, which keeps this above anchor 1 ('only describes rather than instructs'), but the specific steps to execute are missing — anchor 2. | 2 / 5 |
Workflow Clarity | A real sequence is present — FLOW fixes activation check → relay → gather → produce → round → per-utterance branch → closing, and checkpoints are explicit in the contract (every version checked against the reference before it is shown, failed writes recorded with reasons, the record built only after the closing turn's asks are observed). But the sequence is expressed in formalism dense enough that an executor must reverse-engineer it, matching anchor 3 (sequence present, but its checkpoints and steps are implicit in the formalism rather than clearly operationalized); it is not 4 because 'clear sequence' is untenable for this readability. | 3 / 5 |
Progressive Disclosure | The one reference (references/round-composition.md, verified present) is clearly signaled, one level deep, and occasion-bound, but ~550 lines of formal contract are inlined monolithically in SKILL.md where the rubric expects an overview with detail split out. This matches anchor 2 ('content that clearly belongs in separate files is inlined'); it is not 3 because the inlined monolith, not mere organization gaps, is the dominant structure. | 2 / 5 |
Total | 8 / 20 Passed |