Research Critique

Ask yourself:

• What claims lack sufficient evidence?
• Which sources are single-sourced (only one reference)?
• What would break if someone used this in production?
• Are there important caveats missing?

Generate a list of concerns with severity: critical / moderate / minor

Adversarial Reviewer

Ask yourself:

• What alternative explanations exist for the findings?
• What's missing from this analysis that a domain expert would notice?
• Are there biases in the source selection?
• Are any conclusions overstated relative to the evidence?

Generate concerns with severity ratings.

Implementation Engineer

Ask yourself:

• Is this analysis actionable?
• What practical gaps exist for someone trying to use these findings?
• Are version numbers, compatibility info, and setup requirements included?
• Would a developer have enough info to make a decision?

Generate concerns with severity ratings.

Step 3: Gap Analysis

Review all critical concerns. For each one, determine:

• Is this a factual gap (missing data that could be found with more searching)?
• Or a writing issue (the data exists but needs better presentation)?

Step 4: Loop-Back (If Needed)

If there are factual gaps (critical concerns that require more data):

1. Generate 2-3 targeted delta-queries to fill the gaps
2. Read the research-retrieve skill and execute only those specific queries
3. Time-box: spend no more than 3-5 minutes on supplemental searches
4. Maximum 2 loop-backs total

Step 5: Write Results

Write critique.json to the output directory:

{
  "concerns": [
    {
      "persona": "Skeptical Practitioner",
      "concern": "description",
      "severity": "critical|moderate|minor",
      "type": "factual_gap|writing_issue"
    }
  ],
  "loop_backs": 0,
  "delta_queries": [],
  "summary": "Overall assessment of research quality"
}

Step 6: Report Summary

Tell the user:

• Number of concerns by severity
• Whether any loop-backs were triggered
• Overall quality assessment