Content
85%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 tight, highly actionable orchestration skill with an exemplary phased workflow and feedback loops. Its single material defect is that all three referenced detail files are missing from the bundle (and use the nonstandard `reference/` directory name), which both breaks progressive disclosure and leaves the sub-agent prompt template unavailable.
Suggestions
Ship the three referenced files (subagent-prompt.md, sources.md, report.md) in the bundle, and align the directory name with the standard `references/` convention so the markdown links resolve.
Alternatively, inline a minimal version of the sub-agent prompt template in the body, since Phase 3 depends on reproducing it verbatim and nothing in the bundle currently supplies it.
Add a one-line example of the expected findings-file format (claim / quote / URL / date / confidence) so sub-agents produce structurally consistent evidence even without the external template.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and imperative throughout — "Research is parallel; writing is single-point", "These are hard budgets", "raw page content must never enter your context" — with zero padding and no explanation of concepts Claude already knows. Every token earns its place, matching the 5 anchor. | 5 / 5 |
Actionability | Concrete guidance dominates: "Run `date +%Y-%m-%d` via Bash", the exact workspace tree, the depth-mode budget table, and citation rules ([n], [single source], [speculative]). Falls short of 5 because the linchpin sub-agent prompt is deferred to "the locked template in reference/subagent-prompt.md — reproduce it verbatim", and that file is not present in the bundle, leaving a gap in what can actually be executed. | 4 / 5 |
Workflow Clarity | Phases 0–5 are clearly sequenced with explicit validation checkpoints: Phase 4 gap-check against brief.md with delta-query follow-up rounds, "If a sub-agent fails or returns thin results, note it and move on", a deep-mode critique pass, and a resume protocol. This matches the 5 anchor's feedback loops and error-recovery loops. | 5 / 5 |
Progressive Disclosure | References are one level deep and clearly signaled ([reference/subagent-prompt.md], [reference/sources.md], [reference/report.md]), and the inline content is appropriately overview-level. However, none of the three referenced files exist in the bundle (no references/ directory at all), so navigation to the detail content breaks; this falls between the well-structured-but-gapped 3 anchor and the broken-reference case, settling at 3. | 3 / 5 |
Total | 17 / 20 Passed |