Content
77%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 content is a well-sequenced, largely executable phased workflow with strong validation gates and an explicit escalation loop. Its main defects are repetition across the three guidance/deferral sections and a bundle-integrity problem: the rules/ and templates/ files the index leans on are missing, breaking the on-demand loading scheme. The shipped references/*.md files are real and well annotated.
Suggestions
Ship the referenced rules/*.md (layer-decision, spec-first-flow, locator-strategy, token-budget, anti-patterns) and templates/* (install-plan.md, flow.yaml, eas-workflow.yaml, eas-build-profile.json, testid-helper.tsx) files, or inline the critical decision rules — currently ~10 referenced paths dangle and the load-on-demand model cannot function.
Consolidate the overlapping deferral guidance: 'When to use', 'Decision flow at a glance', and 'Composes with' each repeat the web-flows-go-to-e2e-testing and tdd deferrals; one table plus one composes section would reclaim tokens.
Inline a minimal example Maestro flow (launchApp / tapOn id: / assertVisible) in Phase 1 so flow emission is actionable even when templates/flow.yaml is unavailable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and assumes competence: exact bash preflight, terse tables, flag-level defaults, no explanations of known concepts. It fits 4 rather than 5 because the 'When to use', 'Decision flow at a glance', and 'Composes with' sections repeat the same deferrals (web → e2e-testing) several times, and a few rhetorical flourishes ('flows are YAML, not pixel snapshots') could be trimmed. | 4 / 5 |
Actionability | Phase 0 gives runnable read-only commands, Phase 2 prescribes exact flags ('maestro test .maestro/<flow>.yaml', 'record_screen: false', 'retries: 1/2'), and Phase 3 names concrete verification artifacts — mostly executable guidance. It stops short of 5 because the core emission artifacts (sample flow YAML, install plan, setTestId helper) are delegated to templates/ files that are not present in the actual bundle. | 4 / 5 |
Workflow Clarity | The spec → emit-flow → run → heal pipeline is explicitly sequenced with a mandatory Phase 0 gate, a heal-loop cap of 3 attempts, a confidence-based escalation checkpoint ('If confidence is below 90%, escalate'), and a Definition of done checklist. This matches the 5 anchor: explicit validation steps, feedback loops for error recovery, and a checklist for a complex process. | 5 / 5 |
Progressive Disclosure | The SKILL.md is structured as a thin index with well-signaled, annotated, one-level-deep pointers — but scored against the actual bundle, 10 of the ~15 referenced paths (all rules/*.md and templates/* files) do not exist; only the 5 references/*.md files ship. The navigation structure is better organized than the 3 anchor, but the volume of dangling paths is more than the 'minor organization gaps' allowed at 4. | 3 / 5 |
Total | 16 / 20 Passed |