Content
75%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 highly actionable skill body dominated by runnable commands, explicit validation, and a well-structured one-level-deep bundle. The main deductions are verbosity in the Mindset and Anti-Patterns sections (with the staleness point repeated four times) and navigation defects: five dangling evals/ links and a malformed options table.
Suggestions
Remove or fix the Eval Scenarios links to evals/scenario-1..5/task.md — these files do not exist in the bundle, so the section currently breaks navigation.
Trim the Anti-Patterns section by collapsing the WHY and Consequence paragraphs into one line each, and state the registry-staleness point once instead of repeating it across Mindset, When to Use, the second anti-pattern, and Troubleshooting.
Add the missing header row (e.g. | Option | Description | Example | with a separator row) to the Script Options table so it renders as a table.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Most of the body is executable commands rather than explanation, but there is real padding: the Mindset prose ("always at least as stale as the gap since that release"), five anti-patterns each carrying WHY plus a Consequence paragraph, and the registry-staleness point restated in Mindset, When to Use, the second anti-pattern, Troubleshooting, and the References table. It fits "mostly efficient but could be tightened" better than the noticeably-verbose anchor, since none of it explains concepts Claude already knows. | 3 / 5 |
Actionability | Copy-paste ready throughout: script invocations ("scripts/fetch-models.sh --json"), concrete jq filters ("jq '.data[] | select(.policy.state == "enabled")'"), per-command and config-file model switching, and a validation one-liner. The common cases (list, filter, compare, switch, verify) are each covered by a runnable command. | 5 / 5 |
Workflow Clarity | The flow is clear and validation is explicit ("**Validation:** Test the model is active..." with `opencode run "Echo back: model working" && echo "✓ Model active"`), and error recovery is signaled via Troubleshooting pointers. However, the end-to-end sequence is never laid out as an ordered list — the reader assembles it from Usage, Decision Framework, and Switching Models sections — which fits "clear sequence with most checkpoints" rather than the fully sequenced anchor. | 4 / 5 |
Progressive Disclosure | The bundle is appropriately split: two real, substantive one-level-deep references (manual-queries-and-workflows.md, troubleshooting.md) and a script, surfaced through a well-organized References table with a When-to-Use column. It falls short of the top anchor because the Eval Scenarios section links to evals/scenario-1..5/task.md files that do not exist in the bundle, and the Script Options table is missing its header row. | 4 / 5 |
Total | 16 / 20 Passed |