Content
88%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.
An exceptionally dense, actionable verification skill: every section carries operative instructions, the commands are copy-paste ready, and the workflow has explicit validation and feedback loops most skills lack. The two deductions are small — a little justificatory/history prose that could be trimmed, and the fact that the models/ specs and models.tsv the whole document pivots on are missing from the bundle as provided.
Suggestions
Ship the referenced bundle files: the body's run instructions and coverage table point to models/*.tla, models/*.cfg, and models.tsv, none of which are present — scripts/check-models.sh fails immediately because it reads $SKILL/models.tsv.
Trim the 'Where the specs used to live' migration table to one sentence (the old paths only matter for git archaeology) and tighten the intro's rhetorical framing ('the places where a bug does the most harm and tests reach it least') to save ~10 lines of tokens.
The '/statechart' dependency is mentioned only in passing mid-Run section; either move it to a short 'Dependencies' note up front or drop it, so readers know before running whether that skill is required.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body never explains concepts Claude already knows (no 'what is TLA+' or 'what is a lock' padding) and is dense throughout — e.g. the coverage table carries code paths, rules, and harm in one row. Minor trim candidates keep it below a 5: the rhetorical framing 'the places where a bug does the most harm and tests reach it least' and the historical 'Where the specs used to live' migration table are justificatory rather than operative. | 4 / 5 |
Actionability | Fully executable, copy-paste-ready commands cover the common cases: '.cursor/skills/formally-verify-cua-driver/scripts/check-models.sh session' with filter example and env vars (JAVA, TLA2TOOLS, TLC_SECONDS), a by-hand TLC invocation 'java -XX:+UseParallelGC -cp ~/.cache/cua-driver-tla/tla2tools-1.7.4.jar tlc2.TLC -workers 1 -config session.cfg session.tla', and an exact 'git log --oneline cc67e715d1..HEAD -- …' drift command listing every covered file. Counterexample validation gives three concrete checks per replayed step. | 5 / 5 |
Workflow Clarity | The multi-step process is explicitly sequenced with validation checkpoints and feedback loops: match the diff against the coverage table, rerun affected rows, interpret row kinds ('A variant that starts passing means the model lost the step that guard covers. Fix the model, not the row'), replay counterexamples one action at a time against the named functions, reproduce with 'std::sync::Barrier' tests or the verify-cua-driver lane, and 'Do not move it [the sync commit] on a green run alone.' | 5 / 5 |
Progressive Disclosure | The design is 5-level: an overview-first SKILL.md with clearly signaled, one-level-deep references ('Every spec and config lives in this skill's models/ directory', '[models.tsv](models.tsv) lists every config'). But scored against the actual bundle, only scripts/check-models.sh exists — the referenced models/ directory and models.tsv are absent (the script itself reads '$SKILL/models.tsv'), so the primary references dangle; that is more than a minor organization gap, landing between the 3 and 5 anchors. | 4 / 5 |
Total | 18 / 20 Passed |