Content
81%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 highly actionable, well-sequenced multi-phase workflow with concrete prompts, templates, and validation checkpoints. Its main weakness is conciseness: repeated principles and repeated MCP instructions inflate the body without adding information.
Suggestions
Collapse the four principles so they appear once (in Overview) and have the Key Rules section reference them or keep only the operationally distinct rules, removing the near-duplicate restatements.
Extract the repeated MCP poll/save-threadId procedure into a single named block (e.g. a '## Calling the Gemini reviewer' subsection) referenced from Phase 2 and Phase 4 instead of repeating it verbatim.
Consider moving the large round-0 proposal template and the Phase 5 report templates into a references/ file, keeping only the field list inline, to reduce body length and improve progressive disclosure.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The four principles are restated near-verbatim between the Overview and the Key Rules section, and the MCP poll/save-threadId instructions repeat almost verbatim in Phase 2 and Phase 4; the body is mostly efficient but could be meaningfully tightened by deduplicating these. | 3 / 5 |
Actionability | Provides a fully written Gemini review prompt with concrete MCP tool names (mcp__gemini-review__review_start / review_reply_start / review_status), exact scoring dimensions and weights, explicit verdict rules, and complete file templates — copy-paste ready with clear fill-ins. | 5 / 5 |
Workflow Clarity | Phases 0–5 are clearly sequenced with an upfront ASCII flow diagram, an explicit stop condition (score >= SCORE_THRESHOLD, verdict READY, no drift), and validation checkpoints (Anchor Check, Simplicity Check) inside a review→revise feedback loop bounded by MAX_ROUNDS. | 5 / 5 |
Progressive Disclosure | Good section structure with clearly signaled one-level references (shared-references for output protocols, sibling skills under Composing); no bundle files exist. Large inline templates are phase-tied so inlining is defensible, leaving only minor organization gaps. | 4 / 5 |
Total | 17 / 20 Passed |