Content
81%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-designed discovery stub: token-efficient, unambiguous about the first action, with fully executable load/install commands and a verified one-level entry point into a well-organized reference bundle. Weaknesses are minor: a promotional 'Why' section, no inline usage example of actual automation, and a two-level navigation chain where most bundle files are only discoverable after loading core.md.
Suggestions
Add one inline quick-start example (e.g. a 2-3 line navigate-and-screenshot command sequence) so the stub demonstrates a common case without requiring the core.md load first.
Trim the 'Why agent-browser' section to a single line of factual positioning and compress the vendored-version provenance paragraph, which currently spends tokens on advocacy and version-pin details.
In 'Specialized skills' or a short 'Reference files' note, surface the topical bundle files (authentication.md, commands.md, session-management.md, …) directly from SKILL.md so navigation to them does not require loading core.md first.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The stub is lean and assumes competence — no explanation of what browser automation is, commands given without preamble. Minor trim opportunities: the 'Why agent-browser' bullet list ('Fast native Rust CLI, not a Node.js wrapper', 'Works with any AI agent (Cursor, Claude Code, Codex…)') reads partly as advocacy, and the pinned version '0.38.2' plus 'Vendored from…' provenance paragraph are time-sensitive details that could be condensed. This is 'efficient; minor instances' rather than the every-token-earns-its-place level 5. | 4 / 5 |
Actionability | Commands present are fully executable and copy-paste ready: 'npm i -g agent-browser && agent-browser install', 'agent-browser skills get core', 'agent-browser skills get core --full', 'agent-browser skills list', and seven named specialized skills. It stops short of level 5 because the body itself contains no example of actual browser automation usage (no navigate/click/screenshot invocation) — the common cases are entirely delegated to a second load step, a deliberate but real gap in what is written. | 4 / 5 |
Workflow Clarity | This is a short single-purpose dispatcher skill and the simple-skill exception applies: the single action is unambiguous — 'Before running any `agent-browser` command, load the actual workflow content from the CLI' with the exact command, followed by an explicit condition for branching ('Load a specialized skill when the task falls outside browser web pages') with per-skill one-liners. Install → load core → run, with no destructive or batch operations requiring validation checkpoints. | 5 / 5 |
Progressive Disclosure | Good structure: the stub is a genuine overview, points explicitly to references/core.md (verified to exist), and core.md in turn signals the 10 topical reference files plus CLI-served specialized skills. It falls short of level 5 because navigation is two levels deep (SKILL.md → core.md → e.g. commands.md, authentication.md), SKILL.md itself names only 1 of the 11 bundle files, and content is dual-sourced between vendored files and CLI-served copies, adding slight indirection — 'references mostly clear; minor organization gaps'. | 4 / 5 |
Total | 17 / 20 Passed |