Content
80%Reviews 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 command reference with concrete executable examples and a clear numbered core workflow. Its main gaps are the absence of explicit validation/verification checkpoints in the workflow (notably for multi-session state operations) and a monolithic single-file structure that could offload the reference tables to separate bundled files.
Suggestions
Add explicit validation checkpoints to the Core Workflow, e.g. after `click @e4` verify the result with `wait --url` or `snapshot -i` before proceeding, and a 'if the element ref is stale, re-snapshot and retry' feedback loop.
For multi-session/swarm flows, include a verification step after `state save`/`state load` (e.g. confirm auth loaded before scraping) since these are batch-like operations where silent failures are costly.
Move the per-command Quick Reference tables into a bundled reference file (e.g. references/COMMANDS.md) and keep SKILL.md as a concise overview pointing to it, improving progressive disclosure.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and reference-driven: a compact core-workflow block plus command tables and short worked examples, assuming Claude's competence with no padding about what browsers or DOMs are. Every section earns its tokens, e.g. "snapshot -i # -i = interactive elements only". | 3 / 3 |
Actionability | Guidance is fully executable and copy-paste ready, e.g. `agent-browser open <url>`, `agent-browser fill @e3 "text"`, and complete login/form-extraction examples with concrete selectors and refs. No pseudocode; specifics like `wait --url "**/dashboard"` are present. | 3 / 3 |
Workflow Clarity | The Core Workflow numbers its steps (open -> snapshot -i -> interact -> re-snapshot) but there are no explicit validation checkpoints or error-recovery feedback loops, and batch/parallel multi-session operations (e.g. `state save/load auth.json`) lack verification. Per the guideline, missing validation in batch/state operations caps this at 2. | 2 / 3 |
Progressive Disclosure | Content is well-organized into clearly labeled sections, but at ~200 lines it is a single monolithic file with large inline command tables that could be split into separate reference files. No bundle files exist to point to, and the simple-skill carve-out (under 50 lines, no external references needed) does not apply, so it stops at 'some structure, could be better organized'. | 2 / 3 |
Total | 10 / 12 Passed |