Content
60%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 exceptionally actionable — concrete, executable commands with exact flags, output fields, and error-recovery paths — and the workflows are well sequenced with validation checkpoints. However, it is heavily redundant (the provider-override rule and the env-var strings repeat many times) and monolithic, with hundreds of lines of per-provider reference material inlined that belongs in separate bundle files.
Suggestions
Split the per-backend command catalogs (AWS, Blacksmith Testbox, macOS/Windows targets, desktop/WebVNC) and the observability flag reference into references/*.md files, leaving SKILL.md as a concise overview with one-level-deep pointers.
Deduplicate the provider-override policy into a single authoritative statement in the Provider Selection Contract section and remove the restatements scattered through First Checks, Observability, Brokered AWS, and Failure Triage.
Replace the five verbatim repeats of the NODE_OPTIONS/OPENCLAW_* env-var string with one named template (e.g. defined once and referenced as 'the standard test env') to cut significant tokens.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The provider-override policy ('override only when the user explicitly names a backend', 'preserve the resolved provider') is restated more than a dozen times across sections, and the identical NODE_OPTIONS/OPENCLAW_* env-var string is duplicated verbatim in five command blocks. There is no explaining of concepts Claude already knows, which keeps it above a 1, but the repetition is heavy padding for an ~860-line body. | 2 / 5 |
Actionability | Guidance is fully executable throughout: complete shell invocations with flags ('crabbox config show', 'crabbox run --idle-timeout 90m --ttl 240m --timing-json --shell -- ...'), exact JSON summary fields (provider, leaseId, syncDelegated, commandPhases), precise flag semantics, and copy-paste-ready debug and handoff-repair templates covering the common cases. | 5 / 5 |
Workflow Clarity | The document is well sequenced (First Checks, backend gates, observability, E2E lanes, reuse, desktop, failure triage) with a numbered E2E flow and validation checkpoints plus feedback loops (crabbox doctor, one retry with --debug --timing-json, lease cleanup verification, rerun only the failing shard). It falls short of 5 because the overall operational order across the topic sections is implicit rather than one coherent sequence, and checkpoints are scattered. | 4 / 5 |
Progressive Disclosure | There are no bundle files at all (references/, scripts/, assets/ are absent) and the entire per-backend command catalogs, observability flag reference, E2E verification patterns, and failure triage are inlined in one ~860-line SKILL.md. This is content that clearly belongs in separate reference files; the clear section headers keep it above a 1. | 2 / 5 |
Total | 13 / 20 Passed |