Content
48%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 defines a coherent, precisely gated workflow with well-signaled occasion-bound references, but it buries that clarity under a 500-line inline formal contract and heavy cross-section redundancy, and it never concretely exemplifies its central output artifact. It reads as a specification to be studied rather than guidance that can be executed directly.
Suggestions
Move the Lean formal block into a references/ file (e.g. references/formal-contract.md) and keep SKILL.md to the Protocol, Rules, and U-BINDING sections, cutting the most-repeated invariants (the no-selection rule appears ~6 times) to a single authoritative statement.
Add one worked example of a rendered round map — even a compact mock dialogue showing entry relay, a map with frames/branches/origins/unexplored, and the round offer — so the skill's central artifact is concrete rather than described.
Tighten the longest axiom doc comments (e.g. `candidates`, `targets`, `ClosingSupported`, each 10-15 lines of nested subclauses) into short statements of what is judged, moving edge-case detail into the relevant reference file.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The 558-line body is noticeably verbose: the no-selection invariant ("never eliminates, ranks, or scores") is restated roughly six times across the Definition, MORPHISM invariants, the `generate` op, the `candidates` axiom, and the Rules; axiom doc comments run to 15+ lines of nested subclauses; and the Lean block's prose restates what the Protocol and Rules sections say again in natural language. It avoids score 1 only because none of this explains concepts Claude already knows — it is original, if heavily padded, content. | 2 / 5 |
Actionability | There is some genuinely concrete guidance — U-BINDING's five worked invocation examples ("/ideate 'topic or fragments' → U = the argument"), the frame-map options template in references/blank-entry.md, and the explicit offer ordering ("continue first and 'Take this field as it stands' second") — but the skill's central artifact, the rendered round map, is never shown by example, only characterized in one dense 200-word paragraph ("It shows the whole field: the frames and the branches dug under them…"), leaving key execution details abstract. That matches 'some concrete guidance but incomplete' rather than the mostly-executable level of 4. | 3 / 5 |
Workflow Clarity | The multi-step process is clearly sequenced (FLOW's arrow notation, the `start`/`ideate` recursion, and the Protocol section) with explicit checkpoints: the frame-map gate before any candidate ("until the person has chosen frames, no candidate of yours is shown"), every round ending on its offer, and "Before every round's offer, say first what continuing would cost to review and what taking keeps". It falls short of 5 because the same loop is stated in three different notations (arrow flow, Lean definitions, prose) that the reader must reconcile. | 4 / 5 |
Progressive Disclosure | The three reference files exist and are well-signaled with occasion-bound triggers ("read `references/blank-entry.md` before presenting it"; "for `Seeded`, read `references/seeded-entry.md` before the first pass"; "Read `references/round-composition.md` before composing when…") and are one level deep, pointing back to SKILL.md rather than to each other. However, the ~500-line Lean formal contract — content that clearly belongs in a separate reference file — is inlined in full in the always-loaded SKILL.md, matching 'some structure … content that should be separate is inline'. | 3 / 5 |
Total | 12 / 20 Passed |