Content
88%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 tightly written process skill with exemplary sequencing, gates, and checklists, and unusually concrete technique guidance for an instruction-only skill. Its only deductions are light motivational padding and a single-file structure with one external reference whose target file is absent from the bundle.
Suggestions
Trim the motivational editorial lines (e.g., "Be aggressive. Be creative. Refuse to give up.", "debugging superpower") and compress the provenance footnote to a one-line attribution.
Ship the referenced `.agents/skills/diagnose/hitl-loop.template.sh` in the bundle (or move the 10-item loop-construction catalog into a reference file) so the one external reference resolves and the main file stays a lean overview.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with high-signal, imperative guidance and explains nothing Claude already knows, but a few motivational/editorial lines ("Be aggressive. Be creative. Refuse to give up.", "A 2-second deterministic loop is a debugging superpower.", "the bug is 90% fixed") and the long closing provenance paragraph could be trimmed. Efficient with minor over-explanation — the score-4 anchor, clearly above the 'could be tightened' level of 3. | 4 / 5 |
Actionability | For an instruction-only skill the guidance is fully actionable: ten concrete feedback-loop constructions with named tools (Playwright/Puppeteer, curl, bisection harness, differential loop), a concrete hypothesis format template, a specific log-tagging convention (`[DEBUG-a4f2]`), named instrumentation techniques (performance.now(), profiler, query plan), and a template file path for the human-in-the-loop case. Per the rubric's code-vs-instruction note, absence of code is not penalized when guidance is this specific; it does not fall to score 4's 'minor gaps'. | 5 / 5 |
Workflow Clarity | Six clearly sequenced phases with explicit validation gates ("Do not proceed to Phase 2 until you have a loop you believe in", "Do not proceed until you reproduce the bug"), checklists in Phases 2 and 6, and a watched-fail→fix→watched-pass→re-run loop in Phase 5 — matching the score-5 anchor's explicit validation steps, error-recovery feedback loops, and checklists. | 5 / 5 |
Progressive Disclosure | Good structure: a phase-per-section overview with clearly signaled one-level references (`.agents/skills/diagnose/hitl-loop.template.sh` for the last-resort loop, a LICENSE note). However, no bundle files exist alongside this SKILL.md — the referenced template path is not present in the reviewed bundle — and at ~117 lines some content (e.g., the ten loop constructions or the post-mortem hand-off detail) could live in reference files, which is the minor organization gap of the score-4 anchor rather than the clean split of score 5. | 4 / 5 |
Total | 18 / 20 Passed |