Content
56%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 well-structured with clear phases, concrete gating thresholds, and exact output templates, and it correctly defers detail to reference files. But the central 'spawn a different model' mechanic is never made executable, all referenced convention files are absent from the bundle, and version history plus a conformance-test stub inflate the token budget without adding operational value.
Suggestions
Remove the terminal '## Output Format' conformance-test stub (or merge it into the real 'Output format' section) and relocate inline version references (v0.25.1, v0.27.x) to a changelog or deprecated section.
Make the 'Spawn review model' phase executable — either inline the model-selection pairs and the actual spawn invocation, or ship the referenced conventions files (model-routing.md, cross-modal.yaml) inside the skill's references/ directory so the links resolve.
Add an explicit failure-path loop after the Grade step (e.g., 'If ISSUES FOUND: present findings, await user decision, optionally re-review after fixes') so the workflow's validation checkpoint has a recovery path.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient — tight Contract bullets, a gating section with concrete thresholds, and exact output templates — but contains unnecessary padding: version references scattered inline ("v0.25.1 gating", "v0.25.1 extension", "v0.27.x"), a long blockquote disambiguating "gbrain eval cross-modal", and a duplicate terminal "## Output Format" stub whose only stated purpose is to satisfy a conformance test. Matches 'mostly efficient but includes some unnecessary explanation or could be tightened'; not 4 because the version-history material is time-sensitive content outside any old-patterns/deprecated section and the conformance stub is pure redundancy. | 3 / 5 |
Actionability | Concrete elements exist — exact output framing templates with delimiters and field layout, specific invoke/don't-invoke thresholds ("5+ files or 100+ lines", "2+ iterations"), and a numbered refusal-routing chain — but the core mechanic is underspecified: "Spawn review model. Send the work + Contract to a different model" with no commands or mechanism, deferring model selection to conventions/model-routing.md which is not in the bundle. Matches 'some concrete guidance but incomplete... missing key details'; not 4 because the central execution step cannot be carried out from what is written here. | 3 / 5 |
Workflow Clarity | The Phases section gives a clear five-step sequence (Capture → Load Contract → Spawn → Grade → Report) with a built-in checkpoint ("Grade... Pass / fail with specific citations") and the refusal-routing section adds an escalation fallback ("If ALL models in the chain refuse, escalate to the user"), with the user-sovereignty rule closing the loop. Matches 'clear sequence with most checkpoints present; minor validation gaps'; not 5 because there is no explicit feedback loop for what happens when the grade fails (e.g., rework and re-review) — only reporting to the user. | 4 / 5 |
Progressive Disclosure | The body has real section structure and its references (../conventions/cross-modal.yaml, ../conventions/test-before-bulk.md, ../conventions/model-routing.md, skills/testing/SKILL.md) are one level deep and clearly signaled — but none of these files exist in the bundle (no references/, scripts/, or assets/ directories), so the links point outside the skill and cannot be navigated from it, and all substantive content lives inline in a ~175-line monolithic body capped by a redundant conformance-test section stub. Matches 'some structure but could be better organized; references present but not clearly signaling resolvable destinations'. | 3 / 5 |
Total | 13 / 20 Passed |