Content
78%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.
An exceptionally dense, actionable rulebook with real production measurements, executable examples, guards, and a shipping checklist. Its main structural weakness is that everything lives in SKILL.md itself — the trace-tool docs and incident/measurement detail are prime candidates for one-level-deep reference files.
Suggestions
Move the §10 trace-startup CLI flags and run-matrix into a references/ file (e.g. references/trace-startup.md) and keep a two-line pointer plus the exit-code contract in SKILL.md.
Extract the §8/§9 production measurement tables and incident history into a references/measurements.md, leaving the rules and thresholds inline.
Tighten the narrative passages in §8 and §10 (incident storytelling, jank motivation) to one line each so the rules carry the tokens.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and expert-level — it never explains what Postgres, N+1, or CDN caching are, and every rule carries a concrete reason, measurement, or path. A few passages run narrative (the §8 incident recap "Slides startup slowness, Analytics paying startup cost…", §10's opening on jank) and could be trimmed, landing it just below the every-token-earns-its-place anchor. | 4 / 5 |
Actionability | Guidance is copy-paste ready and covers the common cases: a complete Drizzle projection with substr truncation, idempotent CREATE INDEX DDL, the anti-pattern plugin body, a runnable bundle-timing `node -e` command, the full trace-startup invocation with flags, and named guards plus exact file/line pointers (`packages/core/src/deploy/build.ts` `:2384-2410`). | 5 / 5 |
Workflow Clarity | Validation is genuinely present — a pre-ship checklist, guard checks (guard:no-boot-data-work, guard:ssr-cache-shell) with a stated limitation, and the trace tool with documented exit codes 1/2 and run conditions — but the body is a rulebook organized by topic rather than a sequenced workflow with per-task checkpoints, so it sits below the explicit feedback-loop anchor. | 4 / 5 |
Progressive Disclosure | Cross-references to sibling skills and repo files (storing-data, real-time-sync, authentication, org/migrations.ts, trace-startup.mjs) are one level deep and clearly signaled, but there is no references/ directory at all: ~390 lines are inline, including content that would fit separate files — the §10 trace-startup CLI usage and the §8/§9 production measurement tables. | 3 / 5 |
Total | 16 / 20 Passed |