This repository is the version-controlled system of record for DocketDrafter's legal corpora and corpus builders. The public Claude marketplace links here for provenance, while installed skills remain lightweight readers that download published corpus artifacts on first use.
corpora/<name>/references/— generated, reviewable corpus data tracked in ordinary Git.tools/<name>/— source downloaders, builders, and verification tools.tools/publish_corpus.py— deterministic S3 artifact publisher.
The corpora and readers live in separate repositories so Anthropic's marketplace verification does not clone the corpus history.
Set one global environment variable to this repository's root:
export DOCKETDRAFTER_CONTENT_REPO="$HOME/code/docketdrafter-plugin-content"All DocketDrafter corpus readers use $DOCKETDRAFTER_CONTENT_REPO/corpora/<name>/references directly when the variable is available. They do not contact S3 in this mode.
Build and inspect the corpus, commit it, then run:
python tools/publish_corpus.py oh-laws --dry-run
python tools/publish_corpus.py oh-lawsPublishing creates an immutable corpus.tar.xz and manifest.json beneath a version derived from the corpus retrieval date. Artifacts are stored privately in S3 and served through https://public-corpora.docketdrafter.com. Fresh skill environments resolve latest.json once; an installed corpus is never automatically updated.
Metadata changes and promotion of an existing release do not rebuild or upload its archive:
python tools/publish_corpus.py oh-laws --metadata-only
python tools/publish_corpus.py oh-laws --promote 2026-08-13The publisher requires a clean working tree for a real release so the manifest can identify the exact source commit.