Content
73%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 exceptionally actionable with well-sequenced operations and strong validation checkpoints, but it fails structurally: it presents itself as a thin index while the rules/ and templates/ files it links to throughout are missing from the bundle, leaving the actual procedures unreachable. Secondary redundancy around Agent0 handling and repeated inconclusive strings could be tightened. Fixing the broken reference targets is the dominant issue.
Suggestions
Ship the referenced bundle files: create rules/ (spec-format.md, runner.md, preview-url-resolution.md, adversarial.md, out-of-bounds.md, memory.md, spec-sources.md, agent0-runtime.md, preview-auth.md) and templates/ so the 'thin index' links resolve, or inline the missing procedures and stop claiming the detail lives elsewhere.
Consolidate the Agent0 unprepared-host explanation into one place (e.g. rules/agent0-runtime.md) and reference it from run/verify/author instead of restating the scenario and the 'not authored (Agent0 sandbox not prepared: …)' line three times.
State the 'inconclusive: no access path for deployment lookup (pass --url)' rule once (Step 0 or rules/preview-url-resolution.md) and reference it from the run outline and verify, rather than repeating the full string at each mention.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and assumes Claude's competence (no basic-concept explanations), but for a file that declares 'This SKILL.md is a thin index', it carries noticeable repetition: the Agent0 unprepared-host scenario is explained three times (the 'An unprepared Agent0 host' paragraph, the 'author stops with not authored' passage, and its restatement inside 'verify'), and 'inconclusive: no access path for deployment lookup (pass --url)' appears in Step 0, the run outline, and verify. This fits anchor 3 ('mostly efficient but includes some unnecessary explanation or could be tightened') better than anchor 4, whose 'minor instances' understates the redundancy; it is above anchor 2 because most length is genuinely load-bearing operational detail, not padding. | 3 / 5 |
Actionability | Guidance is fully executable: the exact gate command 'node ${CLAUDE_SKILL_DIR}/scripts/is-ui-diff.mjs --base "$(git merge-base origin/HEAD HEAD)"' (script present in scripts/), an mcp-equivalents table ('gh pr view <pr> --json body' → 'mcp__github__pull_request_read with method: "get"'), the complete 80-line on-demand-install bash block with sentinel and budget, exact marker strings ('<!-- ui-verify:v2 -->'), exact report strings, and a decision table keyed on the script's last output line. Specific examples cover the common cases — anchor 5. | 5 / 5 |
Workflow Clarity | Each operation is a numbered sequence with explicit validation checkpoints and feedback loops: author's Step 0 mechanical is-ui-diff gate ('UI_DIFF: no → stop and report not authored'), terminal 'inconclusive' outcomes that must stop without a verdict, the install-sentinel that prevents a second multi-minute attempt, idempotency of verify ('a second verify on the same PR reuses the block'), and hard rules for failure handling ('A missing sandbox browser is NOT RUN (…) … never red'). This matches anchor 5's explicit validation, error-recovery loops, and checklists. | 5 / 5 |
Progressive Disclosure | The body repeatedly delegates its core procedures to files that are not in the bundle: 'Full procedure: rules/runner.md' plus a dozen links to ./rules/*.md (spec-format, preview-url-resolution, runner, adversarial, out-of-bounds, memory, spec-sources, agent0-runtime, preview-auth) and ./templates/*.md — yet the bundle contains only references/ (3 files) and scripts/ (1 file); no rules/ or templates/ directory exists, and even references/adversarial-testing.md opens by pointing to the missing ../rules/adversarial.md. Per the guideline to score against the actual bundle structure, the navigation is broken for the majority of the skill's detail, which is the anchor-2 failure mode (references unusable, needed content unreachable) rather than anchor 3's 'references present but not clearly signaled' — the signals are clear, the targets are absent. | 2 / 5 |
Total | 15 / 20 Passed |