Content
71%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 is an exceptionally actionable command reference with concrete, executable examples for every Crabbox surface and a solid failure-triage section. Its weaknesses are redundancy (the Useful Commands catalog re-lists commands already demonstrated) and the absence of any progressive disclosure: a ~490-line monolith whose provider deep-dives and command catalogs should live in separate reference files.
Suggestions
Split the Hyper-V provider subsection and the 'Useful Commands' catalog into separate reference files (e.g. references/providers-hyperv.md, references/command-reference.md) and link them from short overview sections, so SKILL.md works as a lean index rather than a 490-line monolith.
Remove or drastically trim the 'Useful Commands' section: most entries duplicate commands already shown with context in their home sections, so keeping only commands not demonstrated elsewhere would cut ~30 lines of redundant tokens.
Add a brief top-level workflow (detect config -> verify CLI with doctor -> authenticate -> choose remote surface -> run with preflight/timing -> triage via results/logs -> stop lease) so the overall sequence and its validation gates are explicit rather than implied by section order.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is command-dense with no conceptual padding, but the 40-line 'Useful Commands' catalog duplicates roughly two dozen commands already shown with context (warmup/status/run, job list, pool ready, history/events/logs/results, webvnc, artifacts collect/publish), 'Detect Crabbox' and 'Source Of Truth' overlap on config inspection and doctor, and the Hyper-V section is a dense prose wall. This is more than the 'minor instances' of the level-4 anchor, fitting 'mostly efficient but could be tightened'. | 3 / 5 |
Actionability | Nearly every section gives copy-paste-ready commands with real flags covering the common cases, e.g. 'crabbox run --fresh-pr example-org/my-app#123 --script ./scripts/e2e-smoke.sh' and the warmup -> status --wait -> run -> stop lane. Placeholders like <lease> and <broker-url> are used consistently and appropriately. This matches the fully executable, common-case-covering top anchor. | 5 / 5 |
Workflow Clarity | Per-lane sequences are explicit and ordered (login -> whoami -> doctor; warmup -> status --wait -> run -> stop), validation checkpoints exist ('crabbox doctor', '--preflight', 'job run --dry-run', '--require-artifact'), and the 'Failure Triage' section supplies error-recovery feedback loops. It falls short of a 5 because there is no top-level ordered workflow tying detect -> auth -> choose-surface -> run together, and some checkpoints remain implicit rather than stated as gates. | 4 / 5 |
Progressive Disclosure | Section headers are clear and logically ordered, but this is a ~490-line single file with no bundle: the provider-specific Hyper-V deep-dive and the 'Useful Commands' catalog are exactly the content that belongs in separate reference files. The one external pointer ('See docs/providers/hyperv.md') is a bare inline mention of a file not present in the skill bundle rather than a well-signaled navigable reference. This fits 'some structure but content that should be separate is inline', above the minimal-structure level 2. | 3 / 5 |
Total | 15 / 20 Passed |