Content
56%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 QA-testing core is genuinely good — concrete commands, clear phases, and a rigorous scoring rubric — but it is welded to ~330 lines of auto-generated framework boilerplate that doubles the token cost without helping the task. Dangling file references and duplicated rule numbering add friction. Moving the framework baseline to a shared reference and shipping the referenced template/taxonomy files would lift this substantially.
Suggestions
Move the shared gstack framework baseline (Voice, telemetry onboarding, AskUserQuestion format, Completeness Principle, Contributor Mode, Plan Status Footer — roughly lines 10-340) into a one-level-deep shared reference file, so SKILL.md carries only the QA workflow.
Ship the referenced files with the skill bundle or inline their essentials: 'qa/templates/qa-report-template.md' (the report format the skill repeatedly tells the agent to use) and 'qa/references/issue-taxonomy.md' (the per-page checklist source) are both absent from the bundle.
Fix the structural defects: duplicate rule numbering (rule 11 is 'Show screenshots to the user' and also 'Never fix bugs'), and the dangling 'Find the browse binary:' heading at line 359 that has no content before the next section header.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Roughly half the body (~330 lines: Voice persona, telemetry opt-in dialogs, 'Boil the Lake' promotion, AskUserQuestion format, Contributor Mode, Plan Status Footer) is shared gstack framework content unrelated to QA testing and reads as heavy padding for this skill. The QA-specific portion is dense and command-driven, which keeps it above anchor 1's concept-explanation verbosity. | 2 / 5 |
Actionability | The QA workflow is highly executable — exact browse commands ($B goto, snapshot, fill, click), a parameter table, port-detection snippets, and a fully specified weighted health-score rubric. It falls short of copy-paste-ready 5 because the core deliverable format depends on a template file ('Copy report template from qa/templates/qa-report-template.md') that is not in the bundle. | 4 / 5 |
Workflow Clarity | Setup through Phase 1-6 is clearly sequenced with real checkpoints (NEEDS_SETUP binary check, snapshot -D verification, retry-before-documenting rule, per-page checklist). Not a 5: rule numbering is duplicated (two different rules numbered 11), the 'Find the browse binary:' line dangles with no content before the next header, and the actual workflow is buried after 330 lines of framework boilerplate. | 4 / 5 |
Progressive Disclosure | The document is well-sectioned with inline references (qa/templates/qa-report-template.md, qa/references/issue-taxonomy.md, ETHOS.md), but no bundle files ship with the skill so those references dangle, and ~330 lines of shared framework content that clearly belongs in a one-level-deep reference file are inlined. Structure is present and navigable, so anchor 2's 'minimal structure' does not fit. | 3 / 5 |
Total | 13 / 20 Passed |