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.
A highly actionable, well-sequenced operational guide with executable commands, exact error semantics, and strong validation checkpoints. Its main weaknesses are inlined reference-grade detail (per-command semantics, file protocol) with zero bundle files, and cross-section repetition of the same caveats.
Suggestions
Move the per-command semantics (workflow steps 6-16), the file protocol spec, and the CI/smoke-test patterns into reference files (e.g. references/commands.md, references/file-protocol.md), leaving SKILL.md a lean overview with one-level-deep, clearly signaled pointers.
Deduplicate the WebView/DOM exclusion caveat — state it once in a scope section and drop the repeats in 'What automation cannot verify', the intro, and Notes; same for screenshot determinism, which is explained in three places.
Split the 300-word snapshot.txt bullet into a short header-line summary plus a keyed list (protocol, dispatch_errors, frame_profile, error events) so the file protocol section scans instead of requiring a single dense paragraph.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body assumes competence (no basic-concept padding) and nearly every sentence carries non-obvious operational detail like error names and failure modes. Not 5 because there is real repetition that could be trimmed — WebView/DOM exclusion is stated in four places ('Automation is not browser DOM automation', 'Screenshots of WebView content', the Notes section, and 'WebView DOM interaction is intentionally out of scope') and screenshot determinism is re-explained in the intro, workflow step 12, the Screenshots section, and Notes; not 3 because the padding is redundancy of genuinely operational content, not generic explanation. | 4 / 5 |
Actionability | Everything is copy-paste executable: exact CLI invocations ('native automate assert --timeout-ms 10000 \'4 open\'', 'zig build run -Dplatform=macos -Dautomation=true'), complete JSON payloads, exact error identifiers ('WheelTargetUnknown', 'permission_denied'), and a concrete RuntimeOptions wiring snippet. Not 4 because the examples cover the common cases completely, including failure paths. | 5 / 5 |
Workflow Clarity | The 16-step standard workflow is clearly sequenced with validation checkpoints (wait for 'ready=true', polling asserts, error-event greps), the smoke-test pattern ends in explicit failure conditions ('Fails on timeout or unexpected bridge response'), and the debugging section is an ordered diagnostic ladder with feedback loops (protocol mismatch → rebuild the older side; stale instance → kill and relaunch). Not 4 because checkpoints are explicit at every phase, not implicit. | 5 / 5 |
Progressive Disclosure | The file is a ~250-line monolith with no bundle files at all — per-command semantics (workflow steps 6-16), the file protocol spec, and CI guidance are all inlined where reference files would keep SKILL.md an overview. It matches the anchor 'some structure... content that should be separate is inline': section headers are clear, but nothing is split out. Not 4 because there are no well-signaled references at all — the only external pointer is 'native skills get core --full'; not 2 because the single file is itself well-sectioned and navigable. | 3 / 5 |
Total | 17 / 20 Passed |