Content
85%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 highly actionable, token-efficient CLI reference with a clearly sequenced workflow and explicit re-snapshot checkpoints. Its main defect is that the progressive-disclosure layer is broken in practice: all six references/*.md files and all three templates/*.sh files it points to are absent from the bundle.
Suggestions
Ship the referenced bundle files (references/commands.md, snapshot-refs.md, session-management.md, authentication.md, video-recording.md, proxy-support.md and templates/*.sh) or remove the links — dead reference paths currently break the Deep-Dive Documentation and Ready-to-Use Templates sections.
Drop or merge the 'Form Submission' pattern, which largely repeats the Core Workflow example, to reclaim tokens.
If templates are kept, show their required arguments/expected outputs inline (as the current one-line invocations do) so they remain usable even before the template file is read.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with executable commands and almost free of prose explaining concepts Claude already knows — every section teaches the agent-specific workflow (refs, snapshot lifecycle, sessions) rather than browser basics. It falls short of the 5 anchor mainly because the "Form Submission" pattern substantially repeats the "Core Workflow" example, and the Essential Commands block has some overlap with the pattern sections. It is clearly above the 3 anchor's 'could be tightened' — only minor trimming needed. | 4 / 5 |
Actionability | Nearly the entire body is copy-paste-ready CLI invocations with concrete arguments — `agent-browser fill @e1 "user@example.com"`, `agent-browser wait --url "**/dashboard"`, `agent-browser state save auth.json` — and the common cases (forms, auth, extraction, parallel sessions, iOS) each get a worked example. This matches the 5 anchor (fully executable, common cases covered) with no pseudocode gaps. | 5 / 5 |
Workflow Clarity | The core workflow is explicitly sequenced (Navigate → Snapshot → Interact → Re-snapshot), and a dedicated "Ref Lifecycle (Important)" section plus inline comments ("MUST re-snapshot") give explicit checkpoints after DOM-changing actions; `wait --load networkidle` and `wait --url "**/dashboard"` serve as validation of navigation/login success. This matches the 5 anchor's explicit validation steps and feedback loops; the 4 anchor's 'minor validation gaps' doesn't apply since checkpoints are called out repeatedly. No destructive or batch operations are involved, so no cap applies. | 5 / 5 |
Progressive Disclosure | The structure as written is good — a command overview in SKILL.md with a well-signaled, one-level-deep reference table ('Deep-Dive Documentation' with when-to-use column) plus templates. However, none of the referenced files exist in the bundle: there is no references/ directory (commands.md, snapshot-refs.md, session-management.md, authentication.md, video-recording.md, proxy-support.md) and no templates/ directory (form-automation.sh, etc.). Per the guideline to score against the actual bundle structure, the navigation leads to dead paths, which drops it to the 3 anchor (structure present but the referenced material is not actually delivered); it cannot be a 4/5 when following the links fails. | 3 / 5 |
Total | 17 / 20 Passed |