Content
93%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 a lean, highly actionable map: copy-paste commands for the common loop, a concrete numbered login procedure with an explicit success criterion, and guardrails that are uniformly environment-specific. Its only notable gap is the lack of an inline error-recovery feedback loop — troubleshooting is delegated to the CLI's core guide — which holds workflow clarity at 4.
Suggestions
Add an explicit error-recovery step to the basic loop, e.g. "If a command fails or @eN is not found, run `agent-browser snapshot -i` again and retry with the fresh ref" — this would supply the feedback loop that workflow clarity's top anchor requires.
Name one or two concrete failure statuses `browser_fill_secret` can return and the action each requires, so the "Judge success by the statuses" checkpoint has an explicit recovery path instead of delegating silently to the core guide.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Every line carries environment-specific facts Claude cannot infer ("those flags are discarded", "refs from an older snapshot go stale", "other sessions' browsers run on this machine too", cookie-seed semantics), with zero explanation of concepts Claude already knows. Not 4: there are no instances of over-explanation that could be trimmed. | 5 / 5 |
Actionability | Fully executable, copy-paste-ready commands cover the common case ("agent-browser open https://example.com" through "agent-browser close"), and the login flow gives a numbered procedure with exact tools and parameters ("Call `request_secure_form` with the page's host as `site`", "call `browser_fill_secret` with the handle and the refs of the inputs"). Edge cases are deliberately delegated to `agent-browser skills get core` and `--help` rather than missing, so this fits the top anchor. | 5 / 5 |
Workflow Clarity | Explicit checkpoints are present ("Run `snapshot -i` again after every page change", "Judge success by the statuses `browser_fill_secret` returns", re-snapshot "before acting if the page may have changed"), which places it above anchor 3's implicit/missing checkpoints. However, there is no inline error-recovery feedback loop (e.g., what to do when a command fails or a ref cannot be resolved) — recovery is delegated to the `core` guide — so it falls short of anchor 5's "feedback loops for error recovery". | 4 / 5 |
Progressive Disclosure | The body is explicitly a map ("This file is only the map. Load the real usage guide from the CLI itself") with well-signaled, one-level-deep references, each labeled with its contents ("agent-browser skills get core # workflows, common patterns, troubleshooting", "--full # adds the full command reference", "skills list # specialized guides"). The quick-task loop is inline, and the Rules section is load-bearing guardrails that belongs inline; no bundle files exist to misorganize. This matches the clear-overview anchor with easy navigation. | 5 / 5 |
Total | 19 / 20 Passed |