sn-prepare-citations

Run prepare_citations.py with --report pointing to your Markdown file and --evidence listing all evidence.json files. The script replaces [^source_id] keys with sequential [N] numbers in first-appearance order and appends a numbered bibliography section.

The script normalizes URLs by lowercasing scheme and host, stripping trailing slashes, and dropping fragments. Sources sharing a normalized URL across evidence files merge into a single citation, with the first-seen ID kept as primary.

No, prepare_citations.py uses only the Python 3 standard library: argparse, json, re, dataclasses, pathlib, and urllib.parse. It runs directly with python3 without installing any dependencies.

Unmatched [^key] references are left as-is in the output and reported in the orphan_citations field of the JSON stdout result. The controller guidance says not to deliver the report until a writer or review step fixes or removes those citations.

This is claim-ID leakage, where the writer agent mistakenly used internal evidence claim indexes as citation keys. The script maps each claim ID to its evidence source IDs and substitutes them automatically, reporting resolved and unresolved cases in the claim_id_leakage field.

Yes, pass --no-toc to skip table of contents generation or --no-l0 to skip the key-findings summary box. Both are enabled by default only when an --outline JSON file is provided; without --outline the script runs in legacy citation-only mode.