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, well-guarded skill body with copy-paste commands, explicit cost/auth/exit-code handling, and clearly signaled one-level references. Its weaknesses are meta-scaffolding sections that add tokens without adding guidance and a slightly disorganized References section interrupted by a nested Configuration block.
Suggestions
Delete the meta-scaffolding sections ('Scheduling', 'Goal', 'Intent signature', 'Control-flow features', 'Resource scope' table, 'Preconditions', 'Effects and side effects') — they restate routing and routing-adjacent information already in 'When to use'/'When NOT to use' and the transitions, cutting the body by roughly a third.
Move the '### Configuration' subsection out of '## References' and reunify the split reference list (invocation, execution-protocol, vendor-matrix, prompt-tips, checklist, context-loading) into one annotated list so navigation is unbroken.
Merge 'Dependencies' and 'Tools and instruments', which duplicate the same CLI and resource enumerations, into a single line or fold into the canonical command path section.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The core content (guardrails, canonical commands, exit codes, config migration) is dense and non-obvious, but several sections are padded scaffolding Claude does not need: 'Scheduling/Goal', 'Intent signature', 'Control-flow features', the 'Resource scope' table, and 'Preconditions/Effects' restate information already covered elsewhere. This fits anchor 3 (mostly efficient, some sections that could be tightened) rather than 2, since the padding is a minority of the document and most tokens carry real information. | 3 / 5 |
Actionability | Fully executable guidance throughout: copy-paste-ready commands ('oma image doctor', 'oma image generate "<prompt>" --vendor auto --size auto --quality auto --output json', the --reference variant), a concrete cost threshold ('$0.20'), bypass flags (--yes / OMA_IMAGE_YES=1), env vars, config keys with file paths, and enumerated exit codes. This matches anchor 5 — specific examples cover the common cases. | 5 / 5 |
Workflow Clarity | Entry/Transitions/Failure-and-recovery/Exit gives a clear sequence with explicit error-recovery loops (auth missing, timeout, outdated CLI) and defined exit criteria including partial success. It falls short of anchor 5 because the handoff checklist is delegated to an external file rather than shown, and validation (doctor) is listed as a command rather than woven into the step sequence as an explicit checkpoint. | 4 / 5 |
Progressive Disclosure | References are one level deep and purpose-annotated ('Execution steps (follow for the selected task): resources/execution-protocol.md', 'selected vendor or reference-image operation only' for invocation.md), which is good navigation. It is not a 5 because the '### Configuration' subsection is oddly nested under '## References' and splits the reference list in two, and no bundle files ship alongside SKILL.md in this workspace to verify the referenced paths resolve. | 4 / 5 |
Total | 16 / 20 Passed |