Deep Cite

deep-cite

Purpose

Runs a transparent source-first research workflow: fetch URLs, extract candidate claims, and build citation-ready markdown outputs.

Runbook

1. Prepare a sources.txt file listing one URL per line.
2. Set SOURCE_DIR and OUTPUT_DIR environment variables; create them if absent.
3. Run scripts/fetch-sources.sh sources.txt to download each source into SOURCE_DIR/.
4. Run scripts/extract-claims.sh to parse each downloaded file and emit claims.json.
5. Review claims.json and remove any false positives before proceeding.
6. Run scripts/cite.sh claims.json to produce:
   • OUTPUT_DIR/citations.md with a numbered bibliography
   • OUTPUT_DIR/claims-annotated.md with inline [N] reference markers
7. Print a summary: N sources processed, M claims extracted, P citations generated.

Stop conditions

1. Abort if SOURCE_DIR or OUTPUT_DIR is unset.
2. Abort if every URL fetch fails; individual 4xx/5xx responses should be logged and skipped.
3. Never overwrite OUTPUT_DIR citation files without FORCE=1.
4. Abort if claims.json cannot be written.

Output format

{
  "sources_fetched": 5,
  "claims_extracted": 23,
  "citations_generated": 5,
  "output_dir": "/path/to/output"
}

Example invocations

• SOURCE_DIR=./sources OUTPUT_DIR=./output skills/deep-cite/scripts/run.sh sources.txt
• MOCK_MODE=1 skills/deep-cite/scripts/fetch-sources.sh sources.txt
• "Run deep-cite on these 3 URLs and produce a citations.md file."