Content
63%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 content delivers a genuinely actionable multi-round review workflow with concrete tool names, commands, and polling mechanics. It is dragged down by an off-topic ~25-line scope-limits block inside the Round 1 prompt, duplicated prompt templates, and no progressive disclosure — all templates and prompt text live inline in one long file with no reference bundle.
Suggestions
Move the five Prompt Templates into a single reference file (e.g. references/prompt-templates.md) and keep one representative template inline in Step 2, cutting the SKILL.md body substantially.
Replace the ~25-line SCOPE LIMITS block in the Round 1 prompt with one or two sentences of scope guidance actually relevant to research review (e.g. 'focus on logical gaps and missing experiments; do not manufacture findings'), or move it to a reference file — it currently pads the highest-token-cost part of the skill.
Add an explicit error-recovery checkpoint to the workflow: what to do when review_status never reaches done=true, when a round's response is empty, or when threadId resumption fails, mirroring the existing polling loop.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The workflow steps are written as lean imperatives, but the ~25-line "SCOPE LIMITS" block embedded in the Round 1 prompt discusses threat models, hash/fingerprint schemes, and feature-flag aversion — content unrelated to research review — and the trailing Prompt Templates section restates prompts already given inline. This is 'mostly efficient but includes some unnecessary explanation or could be tightened'; not a 2 because the bulk of the body is functional instruction rather than padding. | 3 / 5 |
Actionability | Concrete, executable guidance dominates: exact MCP tool names (mcp__gemini-review__review_start, review_reply_start, review_status), a runnable install command (`codex mcp add gemini-review -- python3 ~/.codex/mcp-servers/gemini-review/server.py`), polling instructions with waitSeconds/done=true, and copy-ready follow-up phrasings. Minor gaps keep it below 5: the exact call syntax for reply_start is never shown and the Round 1 prompt is a placeholder template ('[Full research context + specific questions]'). | 4 / 5 |
Workflow Clarity | A clearly sequenced 5-step workflow (gather context → round 1 → iterative rounds → convergence → document) with explicit state management (save jobId, poll until done=true, retain threadId) and defined convergence criteria. It sits at anchor 4 rather than 5 because there is no error-recovery checkpoint for reviewer failures/timeouts, and no validation that the saved review document is complete before finishing. | 4 / 5 |
Progressive Disclosure | The body has clear section headers but is a single ~135-line file with no bundle files (references/, scripts/, assets/ do not exist), so everything — the large scope-limits prompt text, the follow-up patterns, and the five prompt templates — is inlined in SKILL.md. This matches 'Some structure but could be better organized; content that should be separate is inline' rather than 4, since the prompt-template library in particular belongs in a one-level-deep reference file. | 3 / 5 |
Total | 14 / 20 Passed |