What it does

Runs the OCaml fetcher + filter in this skill directory:

bin/fetch.ml — fetches (via curl against the GitHub REST/GraphQL API, gh auth token):
1. all merged PRs → merged_prs.json
2. all inline review comments → review_comments.json
3. PR conversation comments + review bodies via GraphQL (connection pagination) → pr_conversations.json
bin/filter.ml — keeps only comments by human authors on merged PRs, drops pure acknowledgements ("LGTM", "Done", "Fixed") and empty bodies → corpus.jsonl with one {kind, pr, author, file, line, created_at, body} per line.

Both steps are idempotent/resumable: existing output files are not re-fetched.

How to run

Requires: gh authenticated (gh auth status), OCaml with yojson (opam), dune, curl.