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 a well-orchestrated, concise phase workflow with genuinely strong validation gates and executable quickstart scripts, but it is structured as a thin index over bundle files that are largely not there: all seven rules/*.md files and the report template are referenced and do not exist, so the progressive-disclosure architecture collapses at every phase beyond the quickstart. Fixing the bundle (or inlining the essential rules) would raise this to a top-tier skill.
Suggestions
Add the missing bundle files — the seven `rules/*.md` files (input-detection, measurement-methodology, action-timing, network-analysis, console-and-errors, flake-diagnosis, confidence-loop) and `templates/analysis-report.md` — or inline their essential content into SKILL.md; currently 8 of 11 referenced paths are dead ends, breaking the entire phase-driven navigation.
Merge the "Workflow" table and the "Required reading by phase" table, which list the same rule files twice, into a single phase table (phase → files → gate) to remove the duplication.
Tighten the "Anti-patterns" section: it promises "full list in references/flake-patterns.md" yet repeats several entries inline — keep the one-liners that frame the skill's philosophy and defer the enumerated traps to the reference file.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and mostly lean — tables for inputs and phases, concrete commands, a checklist — with little explanation of concepts Claude already knows. Not 5: there is real duplication that could be trimmed: the "Workflow" table (phase → rule file → gate) and the "Required reading by phase" table list the same files twice, and the "Confidence-gated iteration" section restates the Phase 4 gate ("≥ 90%", "max 2 deep-dives") already given in the workflow table. Not 3: outside this duplication, every section earns its tokens with skill-specific guidance ("`page.click('text=Save')` waited 4,820ms across 3 attempts"). | 4 / 5 |
Actionability | The quickstart gives copy-paste-ready commands that exist in the bundle: "node <skill_dir>/scripts/trace-extract.mjs <path/to/trace.zip>", "trace-summary.mjs", "trace-diff.mjs", "fetch-gh-run.mjs", each with its expected output described, plus concrete measurement definitions ("`responseEnd - requestStart` (ms)"). Not 5: the core analysis phases (2–5) direct the reader to execute against `rules/*.md` and `templates/analysis-report.md` files that are absent from the bundle, so the workflow cannot actually be followed end-to-end — a significant executable gap beyond the quickstart. | 4 / 5 |
Workflow Clarity | Six named phases with an explicit per-phase gate ("Do not skip a gate"; Phase 0: "Format detected, archive unpacked, `trace.trace` + `trace.network` parseable"), an explicit validation feedback loop ("≥ 90% → proceed; 70–89% → one deeper pass; < 70% → surface the gap to the user"), bounded iteration ("After two deep-dive iterations without reaching 90%, stop"), and a Definition of Done checklist. This matches the anchor-5 pattern of clear sequence, explicit validation, error recovery, and a checklist. Not 4: every checkpoint is stated, none implicit. | 5 / 5 |
Progressive Disclosure | The design intent is exemplary — "the body of `SKILL.md` is a thin orchestrator", "Load on demand — do not preload" — and references are clearly signaled with per-phase loading tables. But scored against the actual bundle: 8 of the 11 referenced paths do not exist. All of `rules/` (input-detection.md, measurement-methodology.md, action-timing.md, network-analysis.md, console-and-errors.md, flake-diagnosis.md, confidence-loop.md) and `templates/analysis-report.md` are missing; only the two `references/*.md` files and four scripts are present. Every phase of the navigation dead-ends at a missing file, so the core content is neither in the body nor reachable — effectively broken navigation rather than imperfect organization. Not 3: the structure that exists is well-signaled, but a majority of the disclosure targets being unresolvable is a more severe failure than "could be better organized". | 2 / 5 |
Total | 15 / 20 Passed |