weakness-scanner

Weakness Scanner

Identify the weakest arguments made across a body of literature. Find logical flaws, data limitations, unsupported claims, and findings contradicted by other work. Your contribution section writes itself after this.

Unlike devils-advocate (which stress-tests YOUR argument), this skill scans OTHER people's work for vulnerabilities. It's how you find the gap your paper fills.

When to Use

• Before writing your contribution section — need to know what's broken in prior work
• Identifying research opportunities — weak arguments = space for new work
• Preparing a rebuttal or response — need to show where existing claims fall short
• Deciding which papers to build on vs. which to challenge

When NOT to Use

• Your own paper — use devils-advocate or the paper-critic agent
• Full peer review — use the referee2-reviewer agent
• Methodological comparison — use method-audit (overlaps, but different focus)

Input

Same corpus inputs: .bib file, PDF directory, topic, or paper list. Works best with 10-20 papers on a focused topic.

Workflow

Phase 1: Corpus Assembly

Same as other corpus skills. Prioritise empirical papers making causal or strong claims — these are most likely to have exploitable weaknesses.

Phase 2: Weakness Extraction

For each paper (read via split-pdf), look for:

1. Logical flaws

   • Non sequiturs — conclusions that don't follow from the evidence
   • Circular reasoning — assuming what they're trying to prove
   • False dichotomies — presenting only two options when more exist
   • Hasty generalisation — drawing broad conclusions from narrow evidence

2. Data limitations

   • Small samples without power analysis
   • Non-representative populations with claims of generalisability
   • Measurement issues (self-report bias, proxy variables)
   • Missing data handled without sensitivity analysis

3. Identification problems

   • Causal claims from observational data without credible identification
   • Omitted variable bias acknowledged but not addressed
   • Reverse causality not ruled out
   • Weak instruments (if IV)