Invocation
- CLI
- REPL
Workflow stages
1
Plan
Before starting, the lead agent outlines the comparison plan: which sources to compare, which dimensions to evaluate, and the expected output structure. The plan is written to
outputs/.plans/<slug>.md and presented to you for confirmation before proceeding.2
Gather source material
When the comparison set is broad, the
researcher subagent gathers source material. If you provide a topic, researchers find the most relevant and contrasting papers. If you provide specific IDs or files, those are used directly.Each source is analyzed independently: claims, results, methodology, and limitations are extracted from each document.3
Build comparison matrix
Claims are aligned across sources, identifying:
- Agreement — where two or more sources make the same claim with supporting evidence
- Disagreement — where sources report contradictory results
- Uncertainty — where sources measure different things or use incompatible evaluation protocols
pi-charts. Method or architecture comparisons use Mermaid diagrams.4
Cite and verify
The
verifier subagent verifies sources and adds inline citations to the final comparison matrix.5
Deliver
Exactly one comparison artifact is saved to
outputs/<slug>-comparison.md. The report ends with a Sources section containing direct URLs for every source used.Outputs
Comparison report structure
- Source summaries — one-paragraph summary of each source’s key contributions
- Agreement matrix — claims supported by multiple sources with citation evidence
- Disagreement matrix — conflicting claims with analysis of why sources diverge
- Methodology differences — how the sources differ in approach, data, and evaluation
- Synthesis — an overall assessment of which claims are well-supported and which remain contested
Subagents used
When to use /compare
Use /compare when:
- You encounter contradictory results in the literature and need to understand whether they reflect genuine disagreement or methodological differences
- Evaluating competing approaches to the same problem
- Writing a related work section that needs to accurately characterize the state of debate
- Deciding which paper’s results to trust when building on prior work
Related
- Literature Review — map the full research landscape on a topic
- Paper Audit — verify a single paper’s claims against its codebase
- Deep Research — deep investigation before comparing specific claims