Content
78%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 dense, highly actionable command reference with a clear three-step workflow and a properly split-out installation reference. Its main weakness is structural: the exhaustive command catalog is inlined in SKILL.md rather than pushed to a references file, and minor flag-variant duplication adds tokens without adding information.
Suggestions
Move the exhaustive per-command listings (Input Automation through Extensions) into a references/commands.md file, keeping the workflow, command-usage pattern, and a clearly signaled pointer in SKILL.md.
Collapse repeated flag-variant lines (e.g., each command duplicated with `--includeSnapshot true`) into a single note that the flag applies to all interaction commands.
Add a brief validation cue in the AI Workflow (e.g., re-snapshot or check console messages after acting) for error recovery.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is almost entirely executable command lines with terse comments and no explanation of concepts Claude already knows (e.g., no "what is a browser devtools" padding). It falls short of 5 due to minor redundancy — many commands are repeated once per flag variant (e.g., `hover 1 "id"` and `hover 1 "id" --includeSnapshot true`) and the AI Workflow section repeats snapshot/click/fill examples that reappear later. | 4 / 5 |
Actionability | Every command is fully executable, copy-paste ready, and presented with inline syntax and flags (e.g., `chrome-devtools navigate_page 1 --url "https://example.com"`, `evaluate_script "(a) => a.innerText" --pageId 1 --args 1_4`), covering the common cases across input, navigation, network, screenshots, and traces. This matches the 'fully executable; copy-paste ready commands covering the common cases' anchor. | 5 / 5 |
Workflow Clarity | The AI Workflow section gives a clear numbered sequence — Execute ("run `chrome-devtools list_pages` to find" the page ID; "do not run `start`/`status`/`stop` before each use"), Inspect (`take_snapshot <pageId>` to get a `<uid>`), Act (click/fill, "State persists across commands") — and the snapshot→uid→act pattern is itself a verification loop. Not 5 because there are no explicit validation checkpoints for risky operations (e.g., verifying a page state after navigation or before uninstall_extension). | 4 / 5 |
Progressive Disclosure | The one bundle reference (references/installation.md) is clearly signaled, correctly scoped as a one-time prerequisite ("Installation is a one-time prerequisite and is **not** part of the regular AI workflow"), and one level deep — but roughly 130 lines of exhaustive per-command API reference (Input Automation through Extensions) are inlined in SKILL.md where they belong in a separate references file, matching the anchor's '[~200 lines of API reference that could be in a separate file]'. The good section headers keep it above 2. | 3 / 5 |
Total | 16 / 20 Passed |