Content
82%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 tight, highly actionable skill body: exact commands, ports, selectors, and a complete happy-path automation snippet with a defined smoke test and expected response. The main gaps are a small executable bug in the snippet's error branches and the absence of explicit error-recovery/retry guidance in the workflow.
Suggestions
Fix the browser_eval snippet's error branches: declare the variable (or use a literal) so '{ ok: false, reason: 'Run task not found', inserted }' does not throw a ReferenceError when the happy path fails.
Add a short retry/poll loop for waiting on the CDP port (e.g., a bounded 'until lsof shows LISTEN' check) and a one-line recovery hint for browser_eval ok:false results to close the workflow's validation loop.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and entirely operational — commands (pnpm dev, nohup, lsof), URLs, ports, a numbered flow, and one working code snippet — with zero explanation of concepts Claude already knows (no CDP or Electron tutorials, no filler). Not a 4 because there is no over-explanation to trim; every section carries non-obvious, project-specific facts. | 5 / 5 |
Actionability | The guidance is largely copy-paste ready: exact shell commands, the CDP URL, target-selection steps, a full browser_eval snippet, and a known-good smoke prompt with its expected response. It falls short of fully executable because the snippet's error branches reference an undeclared shorthand variable 'inserted' ('Run task not found', inserted), which throws a ReferenceError if taken — a minor gap in an otherwise complete example. | 4 / 5 |
Workflow Clarity | The 'Browser Tool Flow' gives a clear numbered 6-step sequence with a verification step ('Confirm the session response by checking document.body.innerText or the current URL'), and background launch includes a port-listen check plus a known-good expected response in Notes. It does not reach 5 because there is no explicit error-recovery loop (e.g., what to do when browser_eval returns ok:false, or how to poll/retry while waiting for the CDP port rather than a single lsof check). | 4 / 5 |
Progressive Disclosure | The body is well-organized into scannable sections (What I Do, Local Dev Setup, Background Launch, Browser Tool Flow, Send A Session, Notes) with everything needed inline and no bundle files to navigate — appropriate since the content is a single coherent procedure, not separable reference material. It is not a 5 because, at ~95 lines, the full browser_eval snippet and smoke-test expectations sit inline in SKILL.md where a small scripts/ or references/ split could keep the overview tighter, and there is no explicit pointer structure to evaluate. | 4 / 5 |
Total | 17 / 20 Passed |