Content
86%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 well-structured, highly actionable body that leans on concrete commands and one-level-deep references. It is tight and navigable; the only marginal improvements are trimming a few RN-recipe sentences and surfacing the RN wait-until-stable checkpoint in the main workflow.
Suggestions
Tighten the RN quick recipe prose (e.g. the perf-overlay sentence) since the full gotcha is already in agent-device-rn.md.
Add the 'wait ~10-13s and screenshot to confirm a stable screen' checkpoint into the main Workflow checklist so the RN timing gate is visible in the body, not only the reference.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Lean and dense with no concept padding; it assumes Claude knows CDP/playwright/RN and every section earns its place, though a few explanatory sentences in the RN quick recipe could be tightened further. | 4 / 5 |
Actionability | Copy-paste-ready bash commands for both backends with concrete testIDs, exact ports, and a parameterized regression runner covering the common desktop/web/RN cases. | 5 / 5 |
Workflow Clarity | A numbered checklist with an explicit evidence gate ('capture AFTER screenshot', 'do not claim fixed without it') and an expected-vs-actual step; the wait-until-stable checkpoint lives in the referenced file rather than the body, leaving a minor gap. | 4 / 5 |
Progressive Disclosure | Overview cleanly points to two real one-level-deep reference files (electron-cdp.md, agent-device-rn.md) that are clearly signaled and cover the detailed material; the bundled script keeps the body short. | 5 / 5 |
Total | 18 / 20 Passed |