Content
77%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 tightly structured two-phase workflow with genuine validation gates, failure guards, and verified one-level-deep references. Main weaknesses are redundancy between inline rules and the anti-patterns section, and pseudocode usage examples that reference a script not present in the bundle.
Suggestions
Consolidate the anti-patterns section: drop entries that restate inline guards (empty-Answer handling, artifact persistence, hypothesis gate, handoff routing) or compress them to one-line pointers, keeping only the WHY/BAD/GOOD patterns that add new content — this would move conciseness toward the 4-5 anchors.
Replace the comment-style usage examples with executable or fill-in-ready probe commands, and either include scripts/run-query-benchmarks.sh in the bundle or change that example to reference a placeholder the user substitutes.
Trim the WHY explanations on anti-patterns that restate in-body mandates (e.g. the empty-Answer guard already appears as a dedicated section) — the rule and its rationale currently appear twice.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The prose is imperative and free of concept padding, but the nine NEVER anti-patterns (~80 lines) largely restate rules already given inline (AskUserQuestion guard, artifact persistence, hypothesis gate, handoff routing), each with an explanatory WHY — more than the minor trimming of the 4 anchor, so it fits the mostly-efficient-but-could-be-tightened 3 anchor. | 3 / 5 |
Actionability | Concrete guidance throughout — exact gate questions, artifact path patterns with collision handling, the 🔬 output template, and a result-routing table — but usage examples are comment-style pseudocode ("# Probe: run both on 1% of traffic for 24h") and one example invokes ./scripts/run-query-benchmarks.sh which does not exist in the bundle. Mostly executable with minor gaps, matching the 4 anchor rather than fully copy-paste-ready 5. | 4 / 5 |
Workflow Clarity | A hard entry gate ("Phase 2 does not start until Phase 1 is complete. No bypass path exists."), pre-registered confirm/refute criteria gated before execution, a feedback loop on gate rejection ("loop back to 1.1–1.4"), an empty-answer failure guard, and a classified exit-gate routing table — explicit validation steps and error-recovery loops matching the 5 anchor. | 5 / 5 |
Progressive Disclosure | All four references/*.md files exist and are one level deep, surfaced in a When-to-Use navigation table; the probe-type taxonomy is correctly offloaded to references/reference.md, and the body stays an operational overview — matching the clear-overview, well-signaled one-level-references 5 anchor. | 5 / 5 |
Total | 17 / 20 Passed |