Claim Extraction

Claim Extraction Format

The claim-extractor agent reads the ingest summary of a repo and produces a markdown document at .research/runs/<timestamp>/claims.md conforming exactly to this schema.

Schema

# Claims extracted from <repo-name>

## Problem
<one to three sentences, in the repo's own framing, of the problem it claims to solve>

## Target user
<one to three sentences describing who the repo says it is for>

## Proposed solution
<two to five sentences: the approach the repo claims to take>

## Quantitative claims

- <verbatim or closely paraphrased claim>  — source: <filename and section>
- ...

## Qualitative claims

- <verbatim adjective-bearing claim>  — source: <filename and section>
- ...

## Differentiation claims

- <what the repo claims to do better than X>  — source: <filename and section>
- ...

## Hedges recorded verbatim

- <any "may", "could", "up to", "in some cases" claim>  — source: <filename and section>
- ...

Rules for the claim-extractor