batch-to-toon

Batch → TOON

Convert many files at once.

Inputs

1. Source directory — absolute path. Non-recursive by default; pass --recursive to walk subdirs.
2. File types — default: .json, .csv, .yaml, .yml. Limit with --ext.
3. Output directory — default: <source>/toon/. Mirrors the input layout.
4. Backend — default: read from plugin config.json (set by setup-tooling); fall back to npx -y @toon-format/toon.

Workflow

1. Glob the source dir for matching files. Skip anything already present in the output dir with the same mtime (idempotent reruns).

2. Create or load a manifest at:

   ${CLAUDE_USER_DATA:-${XDG_DATA_HOME:-$HOME/.local/share}/claude-plugins}/get-toony/state/<batch-id>.json
   

   Each row: input path, type, status (pending / done / failed / skipped-no-savings), original tokens, TOON tokens, saving %, error (if any).

3. For each file:

   • JSON → use json-to-toon skill logic (validate → encode → round-trip).
   • CSV → use csv-to-toon skill logic (CSV → JSON → encode).
   • YAML → parse to JSON first (yq -o=json or python3 -c "import yaml,json; print(json.dumps(yaml.safe_load(open(p))))"), then encode.

4. Round-trip every file. If any fails, mark failed in the manifest and continue — don't abort the batch.

5. Token saving below 5%? Mark skipped-no-savings and leave the original alone. TOON-ifying tiny savings is just churn.

Final report

At the end, print:

• Total files processed / succeeded / failed / skipped.
• Aggregate tokens: total before, total after, absolute saving, % saving.
• Worst-case file (smallest or negative saving) and best-case file.
• Manifest path for inspection.

Resumability

Re-running on the same source dir re-uses the existing manifest. done rows are skipped; pending and failed are retried. --reset forces a clean run.

Failure modes

• YAML files with anchors / merge keys may not round-trip cleanly through JSON. Flag rather than silently lose data.
• CSV files with ragged rows (different column counts per row) won't form a uniform TOON table. Either pad or fail — surface the choice to the user once, then apply consistently across the batch.