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 workflow engineering is excellent: a fully specified, template-driven, bounded review-revise loop with explicit stop conditions, drift handling, and pushback guidance. The weaknesses are length and layout — duplicated instructions and near-identical report templates inflate the file, and everything lives in one monolithic SKILL.md with no bundle files despite several large self-contained template blocks.
Suggestions
Move the large templates (the ~90-line proposal template and the REVIEW_SUMMARY / REFINEMENT_REPORT templates) into references/ files (e.g., references/proposal-template.md, references/report-templates.md) and link them one level deep, which would address both progressive_disclosure and conciseness.
Deduplicate repeated guidance: state the jobId-save/poll/threadId procedure once (Phase 2) and reference it from Phase 4, and consolidate the score-history table and the round-by-round tables that currently appear in both the REVIEW_SUMMARY and REFINEMENT_REPORT templates.
Trim overlapping status reporting: the 'Anchor status / Focus status / Modernity status' bullet set is templated in Steps 5.1, 5.3, and 5.5 — keep it once (e.g., in the user-facing summary) and drop the copies in the file templates.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~670-line body contains noticeable duplication: the jobId/poll instruction appears verbatim in Phase 2 prose, again as a bolded 'CRITICAL' line, and again in Phase 4; the 9-column score-history table is templated three times; and the REVIEW_SUMMARY and REFINEMENT_REPORT templates repeat near-identical round-by-round tables and status blocks. However there is no concept-padding (no explaining things Claude already knows) and the bulk is operational templates, so it sits between the 'padded sections' (2) and 'mostly efficient but could be tightened' (3) anchors — mostly efficient instructions with clear consolidation opportunities. | 3 / 5 |
Actionability | Fully executable throughout: exact MCP tool names (mcp__gemini-review__review_start / review_status / review_reply_start), a complete reviewer prompt with seven dimensions and explicit weights, named constants (MAX_ROUNDS = 5, SCORE_THRESHOLD = 9), exact output paths, and full copy-paste markdown templates for every output file. Matches the 'fully executable, copy-paste ready, covers common cases' anchor. | 5 / 5 |
Workflow Clarity | The phase sequence 0-5 is diagrammed, numbered, and gated by explicit stop conditions (score >= SCORE_THRESHOLD, verdict READY, no unresolved drift, or MAX_ROUNDS), with a loop-back rule from Phase 4 to Phase 3 and an error-recovery loop that classifies reviewer feedback as valid/debatable/drifting and pushes back using the Problem Anchor. Clear sequence with explicit validation checkpoints and feedback loops — the 5 anchor. | 5 / 5 |
Progressive Disclosure | No references/, scripts/, or assets/ files exist — everything is inlined in one ~670-line SKILL.md, including a ~90-line proposal template and two large report templates that clearly belong in separate reference files. The document is well-sectioned with clear headers (above the 2 anchor), and the shared-references links at ../../shared-references/ are one level deep, but content that should be separate is inline and no in-bundle references exist — anchor 3 rather than 4. | 3 / 5 |
Total | 16 / 20 Passed |