Content
81%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-crafted, disciplined skill body: the tool-selection heuristic, the assumptions register, and the counterexample protocol are concrete guidance Claude can execute directly, with genuine validation gates throughout. The main improvements are bundling the ledger-writing details into a reference file and resolving the external "figure-it-out way"/"babysit" pointers that a cold-start agent cannot follow.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes Claude's competence — it never explains what TLA+, Lean, or Kani are, only when each fits — and every section carries actionable rules. Anchor 4 rather than 5 because the opening paragraph ("this skill exists to find that one property, verify it, and leave the verification behind as structure") partially restates the description, and a few rhetorical flourishes could be trimmed. | 4 / 5 |
Actionability | Mostly executable guidance: named run commands ("apalache/TLC, lake build, cargo kani"), an exact output path (".claude/skills/verify-formal-<property>/SKILL.md"), the generated skill's required contents, and concrete ledger imports and row values. Not anchor 5 because step 6's "resolve the package root the figure-it-out way; babysit's ledger step demonstrates the exact invocation" points to context not bundled with this skill, and no exact command-line syntax is given for the verifiers. | 4 / 5 |
Workflow Clarity | Six clearly sequenced steps with explicit validation checkpoints and feedback loops: a violated property triggers minimize-counterexample → check the code → report before touching the model; "formally-verified" is written only after a green rerun; and "If none fits, say so and stop" guards the tool-choice step. This matches the anchor-5 pattern of explicit validation with error-recovery loops. | 5 / 5 |
Progressive Disclosure | Well-organized single-file skill with clear sections (When to use / When not to use / Workflow / Maintenance / Rules) and no nested or buried references. Not anchor 5: the body is ~74 lines (above the sub-50-line simple-skill exception), and step 6's ledger-writing mechanics — plus the dangling pointer to "babysit's ledger step" — are content that arguably belongs in a bundled reference file. | 4 / 5 |
Total | 17 / 20 Passed |