Content
83%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 lean, highly actionable command reference with excellent token efficiency, weakened by the absence of any operational workflow guidance for its destructive and batch commands and by the total lack of bundle files to carry per-command detail.
Suggestions
Add a short safe-editing workflow with an explicit validation checkpoint, e.g. '1. backup is automatic on edit; 2. after any edit run `validate <model>`; 3. on failure run `undo <model>`' — this would lift workflow_clarity past the destructive/batch cap of 3.
Show '--dry-run' for every destructive or batch example (batch, flatten, find-replace, migrate, pack), not just the batch one.
Add a references file (e.g. COMMANDS.md) documenting flags and options for all 42 commands, keeping only the command tables inline in SKILL.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is a lean command catalog: one line per command in dense tables, zero explanation of concepts Claude already knows, and the two trailing usage notes ('All read commands support --json. Edit commands auto-backup.') are load-bearing. This matches anchor 5 — every token earns its place. | 5 / 5 |
Actionability | Each command row pairs a concrete invocation with its purpose, and the Examples section gives ten copy-paste-ready commands with realistic flags ('--target web-sdk', '--dry-run', '--embed', '--fade-in 0.3'), matching anchor 5's 'fully executable; copy-paste ready; specific examples cover the common cases'. | 5 / 5 |
Workflow Clarity | The skill exposes destructive and batch operations (edit commands, find-replace, batch, flatten, migrate, pack) but documents no edit-then-validate feedback loop or recovery sequence — the missing-validation cap for destructive/batch operations applies. Mitigating signals exist ('--dry-run' in a batch example, the undo command, the auto-backup note), which places it at anchor 3 rather than 2, but it cannot exceed 3 without an explicit validate/verify checkpoint. | 3 / 5 |
Progressive Disclosure | Sections are well organized (Read/Edit/Generation/Analysis/Workflow/Other command groups plus Examples) and nothing inlined belongs in a separate file, but at ~100 lines with no bundle files at all, per-command flag/option documentation exists nowhere — the examples reveal flags for only 10 of 42 commands. This matches anchor 4 ('good structure; most content appropriately placed; minor organization gaps') rather than anchor 5's fully split, reference-backed structure. | 4 / 5 |
Total | 17 / 20 Passed |