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 highly actionable, with exact commands, flags, thresholds, exit codes, and a well-specified failure/recovery model, but it pays for this with redundant scheduling boilerplate that restates the same routing rules several times. Its biggest structural defect is that nearly all of the referenced detail files are missing from the bundle, so the progressive-disclosure paths a reader is told to follow do not resolve.
Suggestions
Ship the missing resources/ files (invocation.md, execution-protocol.md, vendor-matrix.md, prompt-tips.md, checklist.md) in the skill bundle or repoint those references to files that actually exist — only references/_shared/core/context-loading.md currently resolves.
Collapse the redundant scheduling sections (Intent signature, When to use / When NOT to use, Control-flow features, Structural Flow, Guardrails) into a single routing section; the pixel-editing boundary and cost-confirmation rule are each restated three or more times.
Add example invocations for the advertised modes that lack them: --vendor all comparison runs, --dry-run cost estimation, and --count/-n usage up to the documented max of 5.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Domain-specific facts (exit codes, $0.20 threshold, env vars, config migration) are dense and non-obvious, but the routing rules are restated 3-4 times across "Intent signature", "When (NOT) to use", "Control-flow features", "Transitions", and "Guardrails" (pixel editing and cost confirmation each appear in multiple sections), matching 'mostly efficient but... could be tightened'. Not 2 because it never explains concepts Claude already knows; not 4 because the scheduling/structural-flow boilerplate is a substantial fraction of the body. | 3 / 5 |
Actionability | Copy-paste-ready commands are given ("oma image generate \"<prompt>\" --vendor auto --size auto --quality auto --output json" and the --reference invocation), plus exact flags (--allow-external-output, --yes, --dry-run), env vars, and enumerated exit codes, matching 'mostly executable guidance... minor gaps'. Not 5 because advertised modes like "--vendor all" comparison, n/count up to 5, and dry-run estimation have no example invocation. | 4 / 5 |
Workflow Clarity | The Entry steps are sequenced (validate -> classify -> check auth/cost/path/count), failure branches name corrective actions ("ask user to run `oma update`", "report vendor-specific authentication requirement"), and validation appears via "oma image doctor" and exit-code-driven error handling, matching 'clear sequence with most checkpoints present'. Not 5 because "Transitions" are prose rules rather than an ordered procedure, and recovery is mostly terminal reporting rather than fix-and-retry loops; not 3 because checkpoints (doctor, cost gate, checklist before handoff) are explicit. | 4 / 5 |
Progressive Disclosure | References are well-signaled with scope conditions ("selected vendor or reference-image operation only", "run before handoff"), but scored against the actual bundle, 5 of the 6 referenced paths (resources/invocation.md, execution-protocol.md, vendor-matrix.md, prompt-tips.md, checklist.md) do not exist — only references/_shared/core/context-loading.md resolves — so navigation to the detailed material is broken. Not 3 because dangling references are a worse navigation defect than the 'not clearly signaled' references at that anchor; not 1 because the body itself is well-sectioned and the reference structure is one level deep in form. | 2 / 5 |
Total | 13 / 20 Passed |